DETAILED ACTION
NOTICE OF PRE-AIA OR AIA STATUS
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
STATUS OF CLAIMS
This action is in response to the Applicant’s arguments and amendments filed on 3/20/2026. Applicant amended claims 1, 29, 33 and 34. Claims 1-2, 4-5, 7, 9, 12-15, 17-27, 29 and 31-34 are pending and are examined below.
FINALITY
Applicant’s request for reconsideration of the finality of the rejection of the last Office Action is persuasive and, therefore, the finality of that action is withdrawn.
RESPONSE TO REMARKS AND ARGUMENTS
In regards to the claim objections, Applicant’s amendments filed on 3/20/2026 obviate the claim objections – accordingly, the claim objections are withdrawn.
In regards to the claim rejections under § 101, Applicant’s amendments filed on 3/20/2026 obviate the claim rejections. Namely, the claims now recite positive vehicle control which necessitates positive actuation of structure; such entails patent-eligible subject matter. Accordingly, the claim objections are withdrawn.
In regards to the claim rejections under § 103, Applicant’s arguments and amendments filed on 3/20/2026 have been fully considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
CLAIM REJECTIONS—35 U.S.C. § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. §§ 102 and 103 (or as subject to pre-AIA 35 U.S.C. §§ 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. § 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claim(s) 1, 2, 4, 5, 9, 12-15, 18, 21 and 32-34 is/are rejected under § 103 as being unpatentable over Rudenko et al. (US20220050469A1; “Rudenko”) in view of Jafari Tafti et al. (US20190204842A1; “Tafti”) and in view of Aoude et al. (“Mobile Agent Trajectory Prediction using Bayesian Nonparametric Reachability Trees”; “Aoude”)1.
As to claim 1, Rudenko discloses a method of generating at least one trajectory in a scenario comprising an agent navigating a mapped area, the method comprising:
receiving an observed state of the agent in an area (Provided is a “machine-learning model which will generate a control trajectory prior based on a current state of the environment including a robot pose …, wherein a pose includes a position and an orientation of the robot or a part of the robot within the environment.” ¶ 15.);
generating a set of multiple trajectory basis elements from the observed state of the agent (“The informed variant of theoretic model predictive control may iteratively evaluate a number of control trajectory samples derived from control trajectory sample prior based on a given distribution at each time step to obtain a further control trajectory sample, wherein the further control trajectory sample is determined depending on a combination of the number of weighted control trajectory samples, wherein the weights are determined based on the costs of each of the number of control trajectory samples.” ¶ 13. “The trajectory prediction model is trained to generate the modelled control trajectory sample based on the actual state of the system.” ¶ 66.);
processing one or more scenario inputs in a neural network (“It may be provided that the data-driven trajectory prediction model comprises a neural network.” ¶ 14.);
generate a set of weights, each weight corresponding to one of the trajectory basis elements (“The data driven trajectory prediction model is trained to estimate the control trajectory prior applying a training data set containing optimized control trajectories for different states of the environment.” ¶ 15.);
generating a trajectory for the agent by weighting each trajectory basis element by its corresponding weights and combining the weighted trajectory basis elements (“The informed variant of theoretic model predictive control may iteratively evaluate a number of control trajectory samples derived from control trajectory sample prior based on a given distribution at each time step to obtain a further control trajectory sample, wherein the further control trajectory sample is determined depending on a combination of the number of weighted control trajectory samples, wherein the weights are determined based on the costs of each of the number of control trajectory samples.” ¶ 13.);
generating control signals for controlling an actor system of the agent based on the trajectory (“The robotic device 1 has a control unit 11 that is configured to perform the subsequently described method and to control movement of the robotic device 1 along a planned trajectory among others.” ¶ 37.); and
controlling the actor system of the agent based on the control signals generated (See at least ¶ 37.).
Rudenko fails to explicitly disclose: processing one or more scenario inputs in a neural network to generate a set of weights, each weight corresponding to one of the trajectory basis elements.
Nevertheless, Tafti teaches: processing one or more scenario inputs in a neural network to generate a set of weights (“The trained neural network generates a navigation trajectory for navigating the vehicle using a cost coefficient determined by the neural network.” Abstract. “The cost function includes different cost components which are pre-defined and different cost coefficients which specify the weights for each cost component. A cost component can be an assigned or calculated energy cost or energy expense of the vehicle for a having collision with other objects on the road, or an assigned or calculated energy cost or energy expense for steering, switching lanes, changing speed, etc.” ¶ 30. “The coefficients αi are included to determine the weight of each cost component in the overall cost of each candidate trajectory.” ¶ 39.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Rudenko to include the feature of: processing one or more scenario inputs in a neural network to generate a set of weights, as taught by Tafti, with a reasonable expectation of success because this feature is useful “to provide an approach to trajectory planning that learns optimal trajectories dynamically for different road scenarios.” (Tafti, ¶ 2.) As mapped above, Rudenko establishes that (1) a neural network processes scenario inputs to inform trajectory generation; and (2) a cost function yields weights corresponding to trajectory basis elements that are combined into a final trajectory. By modifying Rudenko with Tafti’s teaching, one of ordinary skill in the art would have recognized that an enhanced weighting of trajectory basis elements would be obtained by exploiting the advantages offered by a neural network, thereby allowing Rudenko’s system to dynamically learn optimal weighting of trajectory basis elements across different scenarios, as Tafti teaches is desirable.
The combination of Rudenko and Tafti fails to explicitly disclose
receiving an observed state of the agent and map data of the mapped area; and
generating a set of multiple trajectory basis elements from the observed state of the agent based on the map data.
Nevertheless, Aoude teaches:
receiving an observed state of the agent and map data of the mapped area (“Using the intent information, a dynamics model of the target vehicle, and a map of the environment, the trajectory generator outputs a set of predicted trajectories with different probabilities of occurrence.” Section II. at p. 3.)
generating a set of multiple trajectory basis elements from the observed state of the agent based on the map data (“Using the intent information, a dynamics model of the target vehicle, and a map of the environment, the trajectory generator outputs a set of predicted trajectories with different probabilities of occurrence.” Section II. at p. 3.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko and Tafti to include the features of: receiving an observed state of the agent and map data of the mapped area; and generating a set of multiple trajectory basis elements from the observed state of the agent based on the map data, as taught by Aoude, with a reasonable expectation of success because (1) utilizing map data to generate agent trajectories is well known in the art; and (2) these features are useful to “improve the location predictions of agents with uncertain intentions for collision avoidance and conflict detection systems.” (Aoude, Section VI. at p. 12.) While Examiner acknowledges that Aoude’s teaching appears to be directed towards predicting the behavior of non-ego agents as opposed to ego agents as performed in Rudenko, Examiner submits that one of ordinary skill in the art would have found Aoude’s teaching applicable to an ego agent because Aoude, similarly to Rudenko, aims to model a future trajectory of an agent based on weighted trajectory basis elements.
Independent claims 33 and 34 are rejected for at least the same reasons as claim 1 as the claims recite similar subject matter but for minor differences.
As to claim 2, Rudenko discloses: the multiple trajectory basis elements comprise multiple basis paths and weighted basis paths are combined to form a path, the trajectory comprising the path (“The informed variant of theoretic model predictive control may iteratively evaluate a number of control trajectory samples derived from control trajectory sample prior based on a given distribution at each time step to obtain a further control trajectory sample, wherein the further control trajectory sample is determined depending on a combination of the number of weighted control trajectory samples, wherein the weights are determined based on the costs of each of the number of control trajectory samples.” ¶ 13.).
As to claim 4, Rudenko discloses: the multiple trajectory basis elements comprise multiple basis paths and weighted basis paths are combined to form a path (“The informed variant of theoretic model predictive control may iteratively evaluate a number of control trajectory samples derived from control trajectory sample prior based on a given distribution at each time step to obtain a further control trajectory sample, wherein the further control trajectory sample is determined depending on a combination of the number of weighted control trajectory samples, wherein the weights are determined based on the costs of each of the number of control trajectory samples.” ¶ 13.).
The combination of Rudenko and Tafti fails to explicitly disclose:
the multiple trajectory basis elements comprise multiple basis motion profiles and the weighted basis motion profiles are combined to form a motion profile; and
the multiple trajectory basis elements comprise the multiple basis motion profiles and the multiple basis paths, the trajectory comprising the path and the motion profile.
Nevertheless, Aoude teaches:
the multiple trajectory basis elements comprise multiple basis motion profiles and the weighted basis motion profiles are combined to form a motion profile (“We define a motion pattern as a mapping from locations to a distribution over velocities (trajectory derivatives) …. [M]odeling trajectory derivatives is equivalent to modeling trajectories.” Section III. A at p. 3. Continuing, “We define the motion model as a mixture of weighted motion patterns (Eq. (1)) …. This model is the distribution over trajectories we expect for a mobility pattern.” Section III. B. at p. 4.); and
the multiple trajectory basis elements comprise the multiple basis motion profiles and the multiple basis paths, the trajectory comprising the path and the motion profile (“We define a motion pattern as a mapping from locations to a distribution over velocities (trajectory derivatives) …. [M]odeling trajectory derivatives is equivalent to modeling trajectories.” Section III. A at p. 3. Continuing, “We define the motion model as a mixture of weighted motion patterns (Eq. (1)) …. This model is the distribution over trajectories we expect for a mobility pattern.” Section III. B. at p. 4.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko and Tafti to include the features of: the multiple trajectory basis elements comprise multiple basis motion profiles and the weighted basis motion profiles are combined to form a motion profile; and the multiple trajectory basis elements comprise the multiple basis motion profiles and the multiple basis paths, the trajectory comprising the path and the motion profile, as taught by Aoude, with a reasonable expectation of success because (1) utilizing a motion profile to generate agent trajectories is well known in the art as a motion profile is necessarily tied to an agent’s trajectory; and (2) these features are useful to “improve the location predictions of agents with uncertain intentions for collision avoidance and conflict detection systems.” (Aoude, Section VI. at p. 12.) While Examiner acknowledges that Aoude’s teaching appears to be directed towards predicting the behavior of non-ego agents as opposed to ego agents as performed in Rudenko, Examiner submits that one of ordinary skill in the art would have found Aoude’s teaching applicable to an ego agent because Aoude, similarly to Rudenko, aims to model a future trajectory of an agent based on weighted trajectory basis elements.
As to claim 5, the combination Rudenko and Tafti fails to explicitly disclose: the basis motion profiles are generated based on the observed state of the agent and one or more motion profile parameters.
Nevertheless, Aoude teaches: the basis motion profiles are generated based on the observed state of the agent and one or more motion profile parameters (“We define a motion pattern as a mapping from locations to a distribution over velocities (trajectory derivatives) …. [M]odeling trajectory derivatives is equivalent to modeling trajectories.” Section III. A at p. 3. Continuing, “We define the motion model as a mixture of weighted motion patterns (Eq. (1)) …. This model is the distribution over trajectories we expect for a mobility pattern.” Section III. B. at p. 4.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko and Tafti to include the feature of: the basis motion profiles are generated based on the observed state of the agent and one or more motion profile parameters, as taught by Aoude, with a reasonable expectation of success because (1) utilizing a motion profile to generate agent trajectories is well known in the art as a motion profile is necessarily tied to an agent’s trajectory; and (2) this feature is useful to “improve the location predictions of agents with uncertain intentions for collision avoidance and conflict detection systems.” (Aoude, Section VI. at p. 12.) While Examiner acknowledges that Aoude’s teaching appears to be directed towards predicting the behavior of non-ego agents as opposed to ego agents as performed in Rudenko, Examiner submits that one of ordinary skill in the art would have found Aoude’s teaching applicable to an ego agent because Aoude, similarly to Rudenko, aims to model a future trajectory of an agent based on weighted trajectory basis elements.
As to claim 9, Rudenko discloses: scenario inputs comprise one or more of: at least one trajectory prediction (“In step S9 (line 17) of the informed IT-MPC algorithm, the first control data (u*0) is applied to the robotic device 1. The predicted trajectory is updated to the following time step in step S10 (lines 18 to 20) so that (u*0) is set to the former U*1 and so on.” Emphasis added; ¶ 60.).
As to claim 12, Rudenko discloses: wherein the scenario comprises one or more non-ego agents (“The trajectory prediction model which is trained and is to apply a network that can generate robot control trajectories aware of the motions of surrounding dynamic objects and/or individuals.” ¶ 16.).
The combination of Rudenko and Tafti fails to explicitly disclose: wherein the trajectory generation comprises generating a respective predicted trajectory for the one or more non-ego agents wherein the respective predicted trajectory is passed to a planner for planning a trajectory of an ego vehicle.
Nevertheless, Aoude teaches: wherein trajectory generation comprises generating a respective predicted trajectory for the one or more non-ego agents wherein the respective predicted trajectory is passed to a planner for planning a trajectory of an ego vehicle (“A key component for such threat assessment systems is a trajectory prediction algorithm (TPA) that estimates the future states of the target vehicles; these vehicles move in the vicinity of the host vehicle that has the threat assessment system onboard.” Section I. at p. 2. Continuing, the disclosed invention is intended “to be suitable for real-time implementation in collision warning systems or probabilistic path planners for autonomous systems.” Section V. B. at p. 12.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko and Tafti to include the feature of: wherein trajectory generation comprises generating a respective predicted trajectory for the one or more non-ego agents wherein the respective predicted trajectory is passed to a planner for planning a trajectory of an ego vehicle, as taught by Aoude, with a reasonable expectation of success because this feature is useful to “improve the location predictions of agents with uncertain intentions for collision avoidance and conflict detection systems.” (Aoude, Section VI. at p. 12.)
As to claim 13, the combination of Rudenko and Tafti fails to explicitly disclose: generating estimated likelihoods for one or more goals and/or motion profiles of each of one or more non-ego agents, wherein the scenario inputs comprise the estimated likelihoods.
Nevertheless, Aoude teaches: generating estimated likelihoods for one or more motion profiles of each of one or more non-ego agents, wherein the scenario inputs comprise the estimated likelihoods (Through the disclosed algorithm, the motion pattern of a target vehicle may be determined wherein “the probability of the correction motion pattern has approached 1.” Section V. B. at p. 12.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko and Tafti to include the feature of: generating estimated likelihoods for one or more motion profiles of each of one or more non-ego agents, wherein the scenario inputs comprise the estimated likelihoods, as taught by Aoude, with a reasonable expectation of success because this feature is useful to “improve the location predictions of agents with uncertain intentions for collision avoidance and conflict detection systems.” (Aoude, Section VI. at p. 12.)
As to claim 14, Rudenko discloses: the method of claim 1, applied to plan a trajectory for an ego agent, wherein the trajectory generation comprises generating a candidate planned trajectory for the agent (“The informed variant of theoretic model predictive control may iteratively evaluate a number of control trajectory samples derived from control trajectory sample prior based on a given distribution at each time step to obtain a further control trajectory sample, wherein the further control trajectory sample is determined depending on a combination of the number of weighted control trajectory samples, wherein the weights are determined based on the costs of each of the number of control trajectory samples.” ¶ 13.).
As to claim 15, Rudenko discloses: the candidate planned trajectory is used to seed a runtime optimizer that generates a final planned trajectory for the agent (“The goal of IT-MPC is to solve the stochastic optimal control problem.” ¶ 44. “The informed variant of theoretic model predictive control may iteratively evaluate a number of control trajectory samples derived from control trajectory sample prior based on a given distribution at each time step to obtain a further control trajectory sample, wherein the further control trajectory sample is determined depending on a combination of the number of weighted control trajectory samples, wherein the weights are determined based on the costs of each of the number of control trajectory samples.” ¶ 13. NOTE: Here, the IT-MPC analogizes to a runtime optimizer because it is an algorithm which optimizes vehicle trajectory generation in real time. The generation and subsequent analysis of control trajectory samples analogize to seeding said runtime optimizer because they initialize the IT-MPC with candidate solutions from which the optimizer refines into a final trajectory.).
As to claim 18, Rudenko discloses: wherein the trajectory generation further comprises generating one or more spatial uncertainty distributions for the agent, wherein each spatial uncertainty distribution provides a measure of a likelihood of a position of the agent at a given time (“The informed variant of theoretic model predictive control may iteratively evaluate a number of control trajectory samples derived from control trajectory sample prior based on a given distribution at each time step to obtain a further control trajectory sample, wherein the further control trajectory sample is determined depending on a combination of the number of weighted control trajectory samples, wherein the weights are determined based on the costs of each of the number of control trajectory samples.” Emphasis added; ¶ 13.).
As to claim 21, Rudenko discloses: wherein the observed state comprises a position of the agent at an initial timestep (Provided is a “machine-learning model which will generate a control trajectory prior based on a current state of the environment including a robot pose …, wherein a pose includes a position and an orientation of the robot or a part of the robot within the environment.” ¶ 15. “The control trajectory sample prior u* is generated based on the actual state of the system x, the trajectory control set of the last time step u*, initial conditions needed by the network, such as some sensor inputs σ, the robotic device 1 current state φ0, ν0, a covariance metric Σ, terminal costs ϕ, state costs c and the transition model F.” ¶ 63.).
The combination of Rudenko and Tafti fails to explicitly disclose: wherein the observed state comprises a velocity of the agent at an initial timestep.
Nevertheless, Aoude teaches: wherein an observed state comprises a velocity of an agent (“We define a motion pattern as a mapping from locations to a distribution over velocities (trajectory derivatives) …. [M]odeling trajectory derivatives is equivalent to modeling trajectories.” Section III. A at p. 3.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko and Tafti to include the feature of: wherein an observed state comprises a velocity of an agent, as taught by Aoude, with a reasonable expectation of success because it is well known in the art that an initial observed state of an agent may comprise its velocity; such is known to be useful for predicting a trajectory of an agent.
As to claim 32, Rudenko discloses: generating at least one further trajectory basis element wherein the basis element is not based on the map data (“The informed variant of theoretic model predictive control may iteratively evaluate a number of control trajectory samples derived from control trajectory sample prior based on a given distribution at each time step to obtain a further control trajectory sample, wherein the further control trajectory sample is determined depending on a combination of the number of weighted control trajectory samples, wherein the weights are determined based on the costs of each of the number of control trajectory samples.” ¶ 13. “The trajectory prediction model is trained to generate the modelled control trajectory sample based on the actual state of the system.” ¶ 66. Note: The foregoing trajectory samples (basis element(s)) are not necessarily based on map data.).
Claim 7 is/are rejected under § 103 as being unpatentable over Rudenko in view of Tafti and in view of Aoude as applied to claim 2 – further in view of Gao et al. (“VectorNet: Encoding HD Maps and Agent Dynamics from Vectorized Representation”; “Gao”)2.
As to claim 7, the combination of Rudenko and Tafti fails to explicitly disclose: the map comprises a road layout.
Nevertheless, Aoude teaches: the map comprises a road layout (“The map of the environment, typically a road network, is assumed to be available a priori.” Section II. at p. 3.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko and Tafti to include the features of: the map comprises a road layout, as taught by Aoude, with a reasonable expectation of success because utilizing map data, especially map data comprising a road network, to generate agent trajectories is well known in the art.
The combination of Rudenko, Tafti and Aoude fails to explicitly disclose: generating a graph representation of the road layout comprising one or more waypoints, including an observed waypoint and at least one goal waypoint, wherein the multiple basis paths are generated by fitting one or more splines between the observed waypoint and one of the goal waypoints of the graph representation.
Nevertheless, Gao teaches: generating a graph representation of the road layout comprising one or more waypoints, including an observed waypoint and at least one goal waypoint, wherein the multiple basis paths are generated by fitting one or more splines between the observed waypoint and one of the goal waypoints of the graph representation (“For agents, their trajectories are in the form of directed splines with respect to time …. [F]or trajectories, we can just sample key points with a fixed temporal interval (0.1 second), starting from t = 0, and connect them into vectors. Given small enough spatial or temporal intervals, the resulting polylines serve as close approximations of the original map and trajectories.” Section 3.1 at p. 3.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the Rudenko, Tafti and Aoude to include the feature of: generating a graph representation of the road layout comprising one or more waypoints, including an observed waypoint and at least one goal waypoint, wherein the multiple basis paths are generated by fitting one or more splines between the observed waypoint and one of the goal waypoints of the graph representation, as taught by Gao, with a reasonable expectation of success because this feature is useful for reducing computation cost while generating agent trajectories on a road network. (See Gao, Section 5. Conclusion and future work at p. 8.)
Claim(s) 17 and 19 is/are rejected under § 103 as being unpatentable over Rudenko in view of Tafti and in view of Aoude as applied to claim 15 – further in view of Schwalb (US10816978B1; “Schwalb”).
As to claim 17, the combination of Rudenko, Tafti and Aoude fails to explicitly disclose: the method of claim 15, when implemented in a simulation context, wherein the control signals are input to an ego vehicle dynamics model to simulate planned behaviour of the ego agent.
Nevertheless, Schwalb teaches: implementing a simulation context, wherein control signals are input to an ego vehicle dynamics model to simulate planned behaviour of the ego agent (“Wherein the decision module determines the actuator commands based on the AI driver, and simulating behaviors of the ego vehicle object using the actuator commands.” Abstract.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko, Tafti and Aoude to include the feature of: implementing a simulation context, wherein control signals are input to an ego vehicle dynamics model to simulate planned behaviour of the ego agent, as taught by Schwalb, with a reasonable expectation of success because it is well known in the art that simulations provide many benefits for training AI and neural networks in the context of vehicle control (e.g., faster testing speeds, lower computational cost, etc.). (See Schwalb, col. 4, ll. 11-32.)
As to claim 19, Rudenko discloses: wherein generating a trajectory for the agent produces a generated trajectory, and wherein the generated trajectory and the one or more spatial uncertainty distributions are generated for each of a set of timesteps (“The informed variant of theoretic model predictive control may iteratively evaluate a number of control trajectory samples derived from control trajectory sample prior based on a given distribution at each time step to obtain a further control trajectory sample, wherein the further control trajectory sample is determined depending on a combination of the number of weighted control trajectory samples, wherein the weights are determined based on the costs of each of the number of control trajectory samples.” Emphasis added; ¶ 13.).
The combination of Rudenko, Tafti and Aoude fails to explicitly disclose: performing the above in respect to equally-spaced timesteps.
Nevertheless, Schwalb teaches: generating predicted ego vehicle states in equally spaced timesteps (“In a single training set, examples with the same Δt (defined by a known number of time steps according to the table) between the current state and the predicted state can be compared.” - Col. 13. ll. 42-59.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko, Tafti and Aoude to include the feature of: generating predicted ego vehicle states in equally spaced timesteps, as taught by Schwalb, to yield the claim limitation at issue with a reasonable expectation of success because it is well known in the vehicle control art that timesteps are typically set in equally-spaced intervals; Schwalb merely provides the explicit teaching that such is known in the art.
Claim(s) 20 is/are rejected under § 103 as being unpatentable over Rudenko in view of Tafti, in view of Aoude and in view of Schwalb as applied to claim 19 – further in view of Hiruta et al. (US20100324815A1; “Hiruta”).
As to claim 20, the combination of Rudenko, Tafti, Aoude and Schwalb fails to explicitly disclose: wherein the spatial uncertainty distributions are elliptical Gaussian distributions.
Nevertheless, Hiruta teaches: wherein the spatial uncertainty distributions are elliptical Gaussian distributions (“FIG. 8 is a drawing showing a relationship between an own-vehicle position and a long axis and short axis of an error ellipse according to the embodiment.” ¶ 92. “the existence probability of the own-vehicle position is a two-dimensional Gaussian distribution making the own-vehicle absolute position Xc(t) the center of the distribution. Accordingly, depending on a distance from the center of the Gaussian distribution, it is possible to decide the existence probability of the own-vehicle position.” ¶ 93.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko, Tafti, Aoude and Schwalb to include the feature of: wherein the spatial uncertainty distributions are elliptical Gaussian distributions, as taught by Hiruta, with a reasonable expectation of success because this feature is useful for identifying an ego vehicle’s position in an accurate manner. (See Hiruta, ¶¶ 12-13 and 92-93.) Note well that Rudenko also discloses utilizing a “Gaussian” in determining an optimal control distribution at para. [0044]; such provides further motivation that Hiruta’s elliptical Gaussian may be utilized in Rudenko’s determination of a future vehicle trajectory.
Claim 22 is rejected under § 103 as being unpatentable over Rudenko in view of Tafti and in view of Aoude as applied to claim 1 — further in view of Malla et al. (US20220017122A1; “Malla”).
As to claim 22, the combination of Rudenko, Tafti and Aoude fails to explicitly disclose: wherein the trajectory generation comprises generating a trajectory for each of a predefined number of modes, and wherein the method further comprises generating a mode prediction indicating a most likely mode for the agent.
Nevertheless, Malla teaches: wherein trajectory generation comprises generating a trajectory for each of a predefined number of modes, and wherein the method further comprises generating a mode prediction indicating a most likely mode for the agent (“The neural network 108 predicts probabilities pm k, for each agent 202 k and each mode m, such that 133 Σpm k=1 and ranks the predicted trajectories based on the predicted probabilities.” ¶ 47.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko, Tafti and Aoude to include the feature of: wherein trajectory generation comprises generating a trajectory for each of a predefined number of modes, and wherein the method further comprises generating a mode prediction indicating a most likely mode for the agent, as taught by Malla, with a reasonable expectation of success because this feature is useful for enabling a neural network to accurately determine the trajectory an agent. (See Malla, ¶ 47.)
Claims 23-25 are rejected under § 103 as being unpatentable over Rudenko in view of Tafti and in view of Aoude as applied to claim 15 — further in view of Silva et al. (US20210370921A1; “Silva”).
As to claim 23, the combination of Rudenko, Tafti and Aoude fails to explicitly disclose: a collision assessment step comprising computing a likelihood of collision for each agent of the scenario based on the generated trajectories for all agents.
Nevertheless, Silva teaches: a collision assessment step comprising computing a likelihood of collision for each agent of the scenario based on the generated trajectories for all agents (“The vehicle safety system 534 may calculate an overall probability of a collision between the vehicles 102 and 106 by averaging or aggregating the intersection/collision prediction results or probabilities determined in operation 806, over all of the perturbed trajectories of the vehicle 106.” ¶ 119. See also FIG. 8.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko, Tafti and Aoude to include the feature of: a collision assessment step comprising computing a likelihood of collision for each agent of the scenario based on the generated trajectories for all agents, as taught by Silva, with a reasonable expectation of success because this feature is useful to “improve the operation and functioning of autonomous or semi-autonomous vehicles, by more accurately predicting and avoiding potential collisions with other objects moving in the environment.” (Silva, ¶ 16.)
As to claim 24, Rudenko discloses: the trajectory generation further comprises generating one or more spatial uncertainty distributions for the agent, wherein each spatial uncertainty distribution provides a measure of a likelihood of a position of the agent at a given time (“The informed variant of theoretic model predictive control may iteratively evaluate a number of control trajectory samples derived from control trajectory sample prior based on a given distribution at each time step to obtain a further control trajectory sample, wherein the further control trajectory sample is determined depending on a combination of the number of weighted control trajectory samples, wherein the weights are determined based on the costs of each of the number of control trajectory samples.” Emphasis added; ¶ 13.).
The combination of Rudenko, Tafti and Aoude fails to explicitly disclose: the collision assessment step comprises computing a collision probability for the agent by evaluating an overlap between a predicted occupied region of the agent at a given timestep and the generated spatial uncertainty distribution of each further agent at the given timestep.
Nevertheless, Silva teaches: the collision assessment step comprises computing a collision probability for the agent by evaluating an overlap between a predicted occupied region of the agent and the generated spatial uncertainty distribution of each further agent (“The vehicle safety system 534 may calculate an overall probability of a collision between the vehicles 102 and 106 by averaging or aggregating the intersection/collision prediction results or probabilities determined in operation 806, over all of the perturbed trajectories of the vehicle 106.” ¶ 119. See also FIG. 8.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko, Tafti and Aoude to include the feature of: the collision assessment step comprises computing a collision probability for the agent by evaluating an overlap between a predicted occupied region of the agent and the generated spatial uncertainty distribution of each further agent, as taught by Silva, to yield the claim limitation at issue with a reasonable expectation of success because this feature is useful to “improve the operation and functioning of autonomous or semi-autonomous vehicles, by more accurately predicting and avoiding potential collisions with other objects moving in the environment.” (Silva, ¶ 16.) Furthermore, one of ordinary skill in the art would have found it obvious to implement Silva’s processing during Rudenko’s time steps to arrive at the claim limitation as probability calculations are typically performed at a respective time in the vehicle control art.
As to claim 25, the combination of Rudenko, Tafti and Aoude fails to explicitly disclose: providing the collision assessment probabilities and generated trajectories to a further neural network configured to generate trajectories for the agents of the scenario.
Nevertheless, Silva teaches: providing the collision assessment probabilities and generated trajectories to a further neural network configured to generate trajectories for the agents of the scenario (“The components in the memories 518 and 538 (and the memory 554, discussed below) may be implemented as a neural network.” ¶ 82. “The vehicle safety system 534 may calculate an overall probability of a collision between the vehicles 102 and 106 by averaging or aggregating the intersection/collision prediction results or probabilities determined in operation 806, over all of the perturbed trajectories of the vehicle 106.” ¶ 119. See also FIG. 8.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko, Tafti and Aoude to include the feature of: providing the collision assessment probabilities and generated trajectories to a further neural network configured to generate trajectories for the agents of the scenario, as taught by Silva, to yield the claim limitation at issue with a reasonable expectation of success because this feature is useful to “improve the operation and functioning of autonomous or semi-autonomous vehicles, by more accurately predicting and avoiding potential collisions with other objects moving in the environment.” (Silva, ¶ 16.)
Claims 26 and 27 are rejected under § 103 as being unpatentable over Rudenko, in view of Tafti and in view of Aoude as applied to claim 1 — further in view of Casas et al. (US20210272018A1; “Casas”)
As to claim 26, Rudenko discloses:
wherein a neural network is trained for generating the at least one trajectory, the network comprising a set of network parameters (“It may be provided that the data-driven trajectory prediction model comprises a neural network.” ¶ 14. “The data driven trajectory prediction model is trained to estimate the control trajectory prior applying a training data set containing optimized control trajectories for different states of the environment.” ¶ 15. See also ¶ 47 providing discussion as to the network parameters.) the method for training the neural network comprising:
receiving a set of training data comprising a set of training inputs, each input comprising an observed state of an agent (Provided is a “machine-learning model which will generate a control trajectory prior based on a current state of the environment including a robot pose …, wherein a pose includes a position and an orientation of the robot or a part of the robot within the environment.” ¶ 15.);
generating a set of multiple trajectory basis elements from the observed state of the agent (“The informed variant of theoretic model predictive control may iteratively evaluate a number of control trajectory samples derived from control trajectory sample prior based on a given distribution at each time step to obtain a further control trajectory sample, wherein the further control trajectory sample is determined depending on a combination of the number of weighted control trajectory samples, wherein the weights are determined based on the costs of each of the number of control trajectory samples.” ¶ 13. “The trajectory prediction model is trained to generate the modelled control trajectory sample based on the actual state of the system.” ¶ 66.); and
generating a trajectory for the agent by weighting each trajectory basis element by its corresponding weights and combining the weighted trajectory basis elements (“The informed variant of theoretic model predictive control may iteratively evaluate a number of control trajectory samples derived from control trajectory sample prior based on a given distribution at each time step to obtain a further control trajectory sample, wherein the further control trajectory sample is determined depending on a combination of the number of weighted control trajectory samples, wherein the weights are determined based on the costs of each of the number of control trajectory samples.” ¶ 13.).
Rudenko fails to explicitly disclose: processing one or more scenario inputs in a neural network to generate a set of weights, each weight corresponding to one of the trajectory basis elements.
Nevertheless, Tafti teaches: processing one or more scenario inputs in a neural network to generate a set of weights (“The trained neural network generates a navigation trajectory for navigating the vehicle using a cost coefficient determined by the neural network.” Abstract. “The cost function includes different cost components which are pre-defined and different cost coefficients which specify the weights for each cost component. A cost component can be an assigned or calculated energy cost or energy expense of the vehicle for a having collision with other objects on the road, or an assigned or calculated energy cost or energy expense for steering, switching lanes, changing speed, etc.” ¶ 30. “The coefficients αi are included to determine the weight of each cost component in the overall cost of each candidate trajectory.” ¶ 39.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the invention of Rudenko to include the feature of: processing one or more scenario inputs in a neural network to generate a set of weights, as taught by Tafti, with a reasonable expectation of success because this feature is useful “to provide an approach to trajectory planning that learns optimal trajectories dynamically for different road scenarios.” (Tafti, ¶ 2.) As mapped above, Rudenko establishes that (1) a neural network processes scenario inputs to inform trajectory generation; and (2) a cost function yields weights corresponding to trajectory basis elements that are combined into a final trajectory. By modifying Rudenko with Tafti’s teaching, one of ordinary skill in the art would have recognized that an enhanced weighting of trajectory basis elements would be obtained by exploiting the advantages offered by a neural network, thereby allowing Rudenko’s system to dynamically learn optimal weighting of trajectory basis elements across different scenarios, as Tafti teaches is desirable.
The combination of Rudenko and Tafti fails to explicitly disclose: each input an indication of a road layout.
Nevertheless, Aoude teaches: an input comprises a road layout (“The map of the environment, typically a road network, is assumed to be available a priori.” Section II. at p. 3.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko and Tafti to include the feature of: an input comprises a road layout, as taught by Aoude, with a reasonable expectation of success because utilizing map data, especially map data comprising a road network, to generate agent trajectories is well known in the art.
The combination of Rudenko, Tafti and Aoude fails to explicitly disclose:
receiving a set of training data comprising a set of training inputs, each input comprising a set of corresponding ground truth outputs, each ground truth output comprising an actual trajectory taken by the agent from the observed state; and
computing an update to the parameters of the network based on a loss function that penalises deviations between the generated trajectory and the corresponding ground truth output for the agent.
Nevertheless, Casas teaches:
receiving a set of training data comprising a set of training inputs, each input comprising a set of corresponding ground truth outputs, each ground truth output comprising an actual trajectory taken by the agent from the observed state (“In particular, the model trainer 1240 can train a machine-learned model 1135 and/or 1140 based on a set of training data 1245. The training data 1245 can include, for example, real world sensor data and associated ground truth data such as ground truth observed trajectories.” ¶ 149.); and
computing an update to the parameters of the network based on a loss function that penalises deviations between the generated trajectory and the corresponding ground truth output for the agent (“The non-differentiable prior knowledge reward function returns a positive reward when the sample trajectory stays within the reachable area and a ground truth trajectory associated with the object stays within a ground truth reachable area; and the non-differentiable prior knowledge reward function returns a negative reward when the sample trajectory exits the reachable area and the ground truth trajectory associated with the object stays within the ground truth reachable area.” ¶ 103. Note: The foregoing reward function analogizes to the claimed loss function as the two functions both penalize deviations between the generated trajectory and the corresponding ground truth output for the agent.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko, Tafti and Aoude to include the features of: receiving a set of training data comprising a set of training inputs, each input comprising a set of corresponding ground truth outputs, each ground truth output comprising an actual trajectory taken by the agent from the observed state; and computing an update to the parameters of the network based on a loss function that penalises deviations between the generated trajectory and the corresponding ground truth output for the agent, as taught by Casas, with a reasonable expectation of success because these features are useful for training a neural network of an autonomous mobile agent to be better able to generate trajectories and predict behavior. (See Casas, ¶ 78.)
As to claim 27, Rudenko discloses:
wherein the training data comprises historical trajectories taken by agents in past scenarios (“The data driven trajectory prediction model is trained to estimate the control trajectory prior applying a training data set containing optimized control trajectories for different states of the environment.” ¶ 15.), and
wherein the trajectories are generated in training from a past observed state . (“The data driven trajectory prediction model is trained to estimate the control trajectory prior applying a training data set containing optimized control trajectories for different states of the environment.” ¶ 15.)
Claims 29 and 31 are rejected under § 103 as being unpatentable over Rudenko in view of Tafti, in view of Aoude and in view of Casas as applied to claim 26 — further in view of Malla.
As to claim 29, the combination of Rudenko, Tafti and Aoude fails to explicitly disclose: wherein the loss function penalises deviations between the ground truth and the generated trajectory which is most similar to the ground truth only.
Nevertheless, Casas teaches: wherein the loss function penalises deviations between the ground truth and the generated trajectory which is most similar to the ground truth only (“In some implementations, instead of directly optimizing the likelihood of the mixture model, a computing system can heuristically choose the closest matching mode and only apply prediction loss on that mode.” ¶ 55. Continuing, “the non-differentiable prior knowledge reward function returns a positive reward when the sample trajectory stays within the reachable area and a ground truth trajectory associated with the object stays within a ground truth reachable area; and the non-differentiable prior knowledge reward function returns a negative reward when the sample trajectory exits the reachable area and the ground truth trajectory associated with the object stays within the ground truth reachable area.” ¶ 103. Note: The foregoing reward function analogizes to the claimed loss function as the two functions both penalize deviations between the generated trajectory and the corresponding ground truth output for the agent.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko, Tafti and Aoude to include the feature of: wherein the loss function penalises deviations between the ground truth and the generated trajectory which is most similar to the ground truth only, as taught by Casas, with a reasonable expectation of success because this feature is useful for training a neural network of an autonomous mobile agent to be better able to generate trajectories and predict behavior. (See Casas, ¶ 78.)
The combination of Rudenko, Tafti, Aoude and Casas fails to explicitly disclose: a trajectory is generated for each of a predefined number of modes.
Nevertheless, Malla teaches: a trajectory is generated for each of a predefined number of modes (“The neural network 108 predicts probabilities pm k, for each agent 202 k and each mode m, such that 133 Σpm k=1 and ranks the predicted trajectories based on the predicted probabilities.” ¶ 47.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko, Tafti, Aoude and Casas to include the feature of: a trajectory is generated for each of a predefined number of modes, as taught by Malla, with a reasonable expectation of success because this feature is useful for enabling a neural network to accurately determine the trajectory an agent. (See Malla, ¶ 47.)
As to claim 31, the combination of Rudenko, Tafti and Aoude fails to explicitly disclose: wherein the loss function encourages a high probability for the mode whose trajectory is most similar to the ground truth trajectory and a low probability for mode(s) whose trajectories are less similar to the ground truth trajectory.
Nevertheless, Casas teaches: wherein the loss function encourages a high probability for the mode whose trajectory is most similar to the ground truth trajectory and a low probability for mode(s) whose trajectories are less similar to the ground truth trajectory (“In some implementations, instead of directly optimizing the likelihood of the mixture model, a computing system can heuristically choose the closest matching mode and only apply prediction loss on that mode.” ¶ 55. Continuing, “the non-differentiable prior knowledge reward function returns a positive reward when the sample trajectory stays within the reachable area and a ground truth trajectory associated with the object stays within a ground truth reachable area; and the non-differentiable prior knowledge reward function returns a negative reward when the sample trajectory exits the reachable area and the ground truth trajectory associated with the object stays within the ground truth reachable area.” ¶ 103. Note: The foregoing reward function analogizes to the claimed loss function as the two functions both penalize deviations between the generated trajectory and the corresponding ground truth output for the agent.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko, Tafti and Aoude to include the feature of: wherein the loss function encourages a high probability for the mode whose trajectory is most similar to the ground truth trajectory and a low probability for mode(s) whose trajectories are less similar to the ground truth trajectory, as taught by Casas, with a reasonable expectation of success because this feature is useful for training a neural network of an autonomous mobile agent to be better able to generate trajectories and predict behavior. (See Casas, ¶ 78.)
The combination of Rudenko, Tafti, Aoude and Casas fails to explicitly disclose: generate a mode prediction indicating a probability that each mode is an optimal mode.
Nevertheless, Malla teaches: generate a mode prediction indicating a probability that each mode is an optimal mode (“The neural network 108 predicts probabilities pm k, for each agent 202 k and each mode m, such that 133 Σpm k=1 and ranks the predicted trajectories based on the predicted probabilities.” ¶ 47.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the combination of Rudenko, Tafti, Aoude and Casas to include the feature of: generate a mode prediction indicating a probability that each mode is an optimal mode, as taught by Malla, with a reasonable expectation of success because this feature is useful for enabling a neural network to accurately determine the trajectory an agent. (See Malla, ¶ 47.)
CONCLUSION
Any inquiry concerning this communication or earlier communications from the Examiner should be directed to Mario C. Gonzalez whose telephone number is (571) 272-5633. The Examiner can normally be reached M–F, 10:00–6:00 ET.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the Examiner by telephone are unsuccessful, the examiner’s supervisor, Fadey S. Jabr, can be reached on (571) 272-1516. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/MARIO C GONZALEZ/Examiner, Art Unit 3668
1 Aoude, Georges, et al. "Mobile Agent Trajectory Prediction Using Bayesian Nonparametric Reachability Trees." Infotech@Aerospace 2011, 29-31 March, 2011, St. Louis, Missouri, American Institute of Aeronautics and Astronautics, 2011.
Version: Author's final manuscript
2 Gao, Jiyang & Sun, Chen & Shen, Yi & Li, Congcong & Schmid, Cordelia. (2020). VectorNet: Encoding HD Maps and Agent Dynamics From Vectorized Representation. 11522-11530. 10.1109/CVPR42600.2020.01154.
Prosecution Insights