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 .
This action is responsive to the claims dated 5/3/2024.
Claims 1-20 are presented for examination.
Specification
The title of the invention is not descriptive. A new title is required that is clearly indicative of the invention to which the claims are directed.
The following title is suggested: Controlling an Agent Using a Control Neural Network Having a Path Planning Network and a High-Level Action Control Network.
Claim Objections
Claims 8 and 11 are objected to because of the following informalities:
claim 8 recites a set of plurality of trajectory sections, which should read a plurality of trajectory sections or a set of trajectory sections.
claim 11 recites one or computers and one or more storage devices, which should read one or more computers to provide proper antecedent basis for the later-recited the one or more computers.
Appropriate correction is required.
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 (i.e., changing from AIA to pre-AIA ) 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.
Claims 1, 3-6, 11, 13-16, 18, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Shalev-Shwartz et al. (hereinafter Shalev-Shwartz) “Safe, Multi-Agent, Reinforcement Learning for Autonomous Driving” (2016) in view of Vezhnevets et al. (hereinafter Vezhnevets) “FeUdal Networks for Hierarchical Reinforcement Learning” (2017).
Regarding independent claim 1, Shalev-Shwartz teaches a computer-implemented method for controlling an agent interacting with an environment using a control neural network comprising at least (i) a path planning network having a first set of network parameters and (ii) a high-level action control network having a second set of network parameters, the method comprising: at each of a plurality of steps, performing (Shalev-Shwartz: page 1, Abstract, "we decompose the problem into a composition of a Policy for Desires (which is to be learned) and trajectory planning with hard constraints (which is not learned)"; page 9, Section 6, "Each policy function, associated with every node of the option graph, is represented by a neural network with three fully connected hidden layers"; the trained deep-network driving policy is the control neural network (a control neural network) that controls a host vehicle (an agent) interacting with a driving environment (an environment), and that policy includes a learned Policy for Desires (a high-level action control network) governed by its own trainable network weights (a second set of network parameters); Shalev-Shwartz thus teaches the general two-part decomposition and the learned high-level action control network):
obtaining an observation of the environment at a current step (Shalev-Shwartz: page 2, "which represents the sensing state of the system, i.e., the environmental model"; page 9, Section 6, "Each agent also observes the location, velocity, and heading of every other car which is within 100 meters away from it"; at each decision step the host vehicle obtains its sensed environmental model (an observation of the environment at a current step));
processing a second input comprising data characterizing (i) the observation at the current step and using the high-level action control network, according to values of the second set of network parameters, to generate a high-level action control output specifying one or more planned actions for the agent (Shalev-Shwartz: page 8, Section 5, "we need to decide whether to change lane (to left or right side) or to stay in lane"; the learned Policy for Desires (the high-level action control network) processes the sensed state (the observation at the current step) and outputs high-level maneuver decisions such as lane changes (a high-level action control output specifying one or more planned actions for the agent) selected according to the trained node weights (values of the second set of network parameters));
generating, using the planned actions generated by the high-level action control network, control data for the agent for the current step (Shalev-Shwartz: page 6, Section 4, "maps the Desires into a trajectory (which determines how the car should move in a short range)"; the selected Desires (the planned actions) are converted into a driving trajectory that determines how the host vehicle moves (control data for the agent for the current step)); and controlling the agent using the control data to interact with the environment (Shalev-Shwartz: page 6, Section 4, "a trajectory that minimizes the aforementioned cost subject to hard constraints on functional safety"; page 9, Section 6, "we used a simulator (initialized using imitation learning) with self-play enhancement"; the computed driving trajectory (the control data) drives the host vehicle (the agent) so that it moves within and reacts to the surrounding driving environment (to interact with the environment)).
Shalev-Shwartz does not expressly teach processing a first input comprising data characterizing the observation at the current step using the path planning network, according to values of the first set of network parameters, to generate a path planning output specifying a planned trajectory for the agent, or that the second input processed by the high-level action control network comprises data characterizing (ii) the planned trajectory generated by the path planning network.
However, Vezhnevets teaches processing a first input comprising data characterizing the observation at the current step using the path planning network, according to values of the first set of network parameters, to generate a path planning output specifying a planned trajectory for the agent (Vezhnevets: page 2, "The Manager and the Worker share a perceptual module which takes an observation from the environment"; page 3, "we ought to end up taking advantageous trajectories through state-space"; page 3, "an advantageous direction in the latent state space"; the Manager network (the path planning network) processes the shared representation of the environment observation and outputs a goal that specifies an advantageous direction, that is, a planned trajectory through the agent's state space (a path planning output specifying a planned trajectory for the agent); under the broadest reasonable interpretation adopted above, a planned trajectory reads on such a directional plan or route through the agent's state space and need not be a metric path through physical space, so the Manager's goal is a path planning output specifying a planned trajectory for the agent, and this output is governed by the Manager's trained weights (values of the first set of network parameters)), and further teaches that the high-level action control network processes as an input (ii) the planned trajectory generated by the path planning network (Vezhnevets: page 2, "conditioned on external observation, its own state, and the Managers goal"; page 1, Abstract, "The Manager operates at a lower temporal resolution and sets abstract goals which are conveyed to and enacted by the Worker"; the Worker network (the high-level action control network) produces its actions conditioned on both the observation and the goal supplied by the Manager (the planned trajectory generated by the path planning network)).
Because Shalev-Shwartz and Vezhnevets are analogous art in the same field of endeavor, namely the hierarchical reinforcement-learning control of an agent that senses an environment and selects actions, and both are reasonably pertinent to the same problem of decomposing long-horizon sequential control into a higher-level planning component and a subordinate action-selection component, accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to implement the two-part driving policy of Shalev-Shwartz as the fully-differentiable two-network hierarchy of Vezhnevets, so that a path planning network first processes the observation to generate a planned trajectory and the high-level action control network then selects the planned actions conditioned on both the observation and that planned trajectory, with a reasonable expectation of success, to teach processing a first input ... using the path planning network ... to generate a path planning output specifying a planned trajectory for the agent and processing (ii) the planned trajectory generated by the path planning network at the high-level action control network. This modification would have been motivated by the desire to make long-horizon credit assignment more tractable and to structure the agent's behaviour into coherent, temporally extended sub-policies, as taught by Vezhnevets (Vezhnevets: page 1, Abstract, "facilitating very long timescale credit assignment"), which is the same long-horizon variance problem that Shalev-Shwartz addresses through temporal abstraction.
Regarding dependent claim 3, Shalev-Shwartz, in view of Vezhnevets, teach the method of claim 1, wherein the environment is a simulated environment, and obtaining the observation of the environment at the current step comprises: performing a simulation of the agent interacting with the simulated environment to generate a simulated observation (Shalev-Shwartz: page 9, Section 6, "we used a simulator (initialized using imitation learning) with self-play enhancement"; the host vehicle is run in a driving simulator (a simulated environment) whose produced state is fed back to the policy (a simulated observation)).
Regarding dependent claim 4, Shalev-Shwartz, in view of Vezhnevets, teach the method of claim 3, wherein the simulated environment comprises one or more other simulated agents that are controlled by one or more respective control models (Shalev-Shwartz: page 9, Section 6, "Set A was used as reference players while set B was used for the policy gradient learning process"; the simulator contains other simulated vehicles (one or more other simulated agents) that are each driven by their own policy (one or more respective control models)).
Regarding dependent claim 5, Shalev-Shwartz, in view of Vezhnevets, teach the method of claim 1, wherein the agent is one of: a vehicle, a vehicle convoy, a pedestrian, or a robotic agent (Shalev-Shwartz: page 7, Section 4, "the desired target speed of the host vehicle"; the controlled host vehicle is a vehicle (a vehicle), and this recitation is an alternative grouping for which teaching one listed member is sufficient).
Regarding dependent claim 6, Shalev-Shwartz, in view of Vezhnevets, teach the method of claim 1, wherein the one or more planned actions specified by the high-level action control output comprise: a turn, a lane switch, or a change of speed (Shalev-Shwartz: page 8, Section 5, "we need to decide whether to change lane (to left or right side) or to stay in lane"; the Policy for Desires selects a change of lane (a lane switch), and this recitation is an alternative grouping for which teaching one listed member is sufficient).
Regarding independent claim 11, Shalev-Shwartz teaches a system for controlling an agent interacting with an environment using a control neural network comprising at least (i) a path planning network having a first set of network parameters and (ii) a high-level action control network having a second set of network parameters, the system comprising: one or more computers and one or more storage devices storing instructions that when executed by the one or more computers cause the one or more computers to perform operations (Shalev-Shwartz: page 4, Section 2, "is implemented by a deep layered network"; page 9, Section 6, "Each policy function, associated with every node of the option graph, is represented by a neural network with three fully connected hidden layers"; the deep-network driving policy executes as programmed instructions on computing hardware (one or more computers and one or more storage devices storing instructions)). The remaining substantive limitations of claim 11 recite the system-form counterparts of the operations of method claim 1 and are taught by Shalev-Shwartz in view of Vezhnevets for the same reasons set forth above for claim 1. The motivation to combine Shalev-Shwartz and Vezhnevets is the same as set forth for claim 1.
Regarding dependent claims 13-16, the rejections of claims 3-6, respectively, are applied in the same manner to the corresponding system form recitations of claims 13-16.
Regarding independent claim 18, Shalev-Shwartz teaches one or more non-transitory computer-readable storage media storing instructions that when executed by one or more computers cause the one or more computers to perform operations for controlling an agent interacting with an environment using a control neural network comprising at least (i) a path planning network having a first set of network parameters and (ii) a high-level action control network having a second set of network parameters (Shalev-Shwartz: page 4, Section 2, "is implemented by a deep layered network"; the trained driving policy is stored and executed as machine-readable instructions (one or more non-transitory computer-readable storage media storing instructions)). The remaining substantive limitations of claim 18 recite the media form counterparts of the operations of method claim 1 and are taught by Shalev-Shwartz in view of Vezhnevets for the same reasons set forth above for claim 1. The motivation to combine Shalev-Shwartz and Vezhnevets is the same as set forth for claim 1.
Regarding dependent claim 20, the rejection of claim 3 is applied in the same manner to the corresponding media-form recitation of claim 20.
Claims 2, 12, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Shalev-Shwartz, in view of Vezhnevets, as applied in the rejections of claims 1, 11, and 18, respectively above, and further in view of Codevilla et al. (hereinafter Codevilla) “End-to-end Driving via Conditional Imitation Learning” (Mar. 2, 2018).
Regarding dependent claim 2, Shalev-Shwartz, in view of Vezhnevets, teach the method of claim 1 as set forth above.
Shalev-Shwartz and Vezhnevets do not expressly teach wherein: the control neural network further comprises a low-level action control network having a third set of network parameters; generating the control data comprises: processing a third input comprising data characterizing (i) the observation at the current step and (ii) the one or more planned actions specified by the high-level action control output using the low-level action control network, according to values of the third set of network parameters, to generate a low-level action control output specifying one or more control signals for the agent; and controlling the agent using the control data to interact with the environment comprises: applying the control signals specified in the low-level action control output to the agent.
However, Codevilla teaches the control neural network further comprises a low-level action control network having a third set of network parameters; generating the control data comprises: processing a third input comprising data characterizing (i) the observation at the current step and (ii) the one or more planned actions specified by the high-level action control output using the low-level action control network, according to values of the third set of network parameters, to generate a low-level action control output specifying one or more control signals for the agent (Codevilla: page 3, Section IV, "the action space is continuous and two-dimensional: steering angle and acceleration"; page 3, "The network takes the command as an input, alongside the image and the measurements"; page 3, "These three inputs are processed independently by three modules"; page 4, Section IV, "The branches Ai are forced to learn sub-policies that correspond to different commands"; a further command-conditioned deep network (a low-level action control network) processes the observation together with the high-level command (the one or more planned actions specified by the high-level action control output) and outputs the steering-angle and acceleration values (a low-level action control output specifying one or more control signals) under its own trained weights (a third set of network parameters)), and further teaches controlling the agent using the control data to interact with the environment comprises: applying the control signals specified in the low-level action control output to the agent (Codevilla: page 6, Section V, "The predicted control is forwarded to the Pixhawk, which controls the car accordingly by sending the appropriate PWM signals to the speed controller and steering servo"; the predicted steering and acceleration values (the control signals) are applied to actuate the vehicle (applying the control signals specified in the low-level action control output to the agent)).
Because Shalev-Shwartz, Vezhnevets, and Codevilla are analogous art in the same field of endeavor of learned control of a vehicle agent, and each is reasonably pertinent to the problem of translating high-level driving decisions into executable low-level control, accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to add to the hierarchy of Shalev-Shwartz and Vezhnevets a low-level action control network as taught by Codevilla, conditioned on the observation and the high-level planned actions and outputting the control signals that are applied to the agent, with a reasonable expectation of success, to teach the low-level action control network limitation of claim 2. This modification would have been motivated by the desire to provide a controllable network that handles sensorimotor coordination while continuing to follow the higher-level decisions, as taught by Codevilla (Codevilla: page 1, Abstract, "the learned driving policy functions as a chauffeur that handles sensorimotor coordination but continues to respond to navigational commands").
Regarding dependent claims 12 and 19, the rejection of claim 2 is applied in the same manner to the corresponding system form and media form recitations of claims 12 and 19, respectively.
Claims 7, 10, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Shalev-Shwartz, in view of Vezhnevets, as applied in the rejections of claims 1, 11, and 18, respectively above, and further in view of Kuefler et al. (hereinafter Kuefler) “Imitating Driver Behavior with Generative Adversarial Networks” (2017) and Xu et al. (hereinafter Xu) “End-to-end Learning of Driving Models from Large-scale Video Datasets” (2017).
Regarding dependent claim 7, Shalev-Shwartz, in view of Vezhnevets, teach the method of claim 1, and further teach that the control neural network is trained by reinforcement learning (Shalev-Shwartz: page 1, Abstract, "we apply deep reinforcement learning to the problem of forming long term driving strategies").
Shalev-Shwartz and Vezhnevets do not expressly teach wherein the control neural network has been trained in a reinforcement learning process using a sequence of frames of image data recorded from a real-world scene.
However, Kuefler teaches training a driving-policy network by a reinforcement-learning process from real-world driving demonstrations (Kuefler: page 1, Section I, "we apply GAIL to the task of modeling human highway driving behavior"; page 1, Section I, "expert demonstrations are given by real-world driver trajectories included in the NGSIM dataset"; the driving policy (the control neural network) is trained by a generative-adversarial reinforcement-learning process (a reinforcement learning process) from records of real highway driving (a real-world scene)), and Xu teaches that such real-world driving records are a sequence of image frames (Xu: page 1, Abstract, "learning a generic vehicle motion model from large scale crowd-sourced video data"; page 5, Section 4, "The Berkeley DeepDrive Video dataset (BDDV) is a dataset comprised of real driving videos"; the training data is a sequence of frames of real driving video (a sequence of frames of image data recorded from a real-world scene)).
Because Shalev-Shwartz, Vezhnevets, Kuefler, and Xu are analogous art in the same field of endeavor of learned control of a vehicle agent, and each is reasonably pertinent to the problem of obtaining realistic driving behaviour, accordingly, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to train the reinforcement-learning control network of Shalev-Shwartz and Vezhnevets using a sequence of real-world driving video frames as taught by Kuefler and Xu, with a reasonable expectation of success, to teach wherein the control neural network has been trained in a reinforcement learning process using a sequence of frames of image data recorded from a real-world scene. This modification would have been motivated by the desire to learn robust, realistic behaviour that generalizes across diverse real-world conditions, as taught by Xu (Xu: page 1, Abstract, "Robust perception-action models should be learned from" training data with diverse visual appearances and realistic behaviors).
Regarding dependent claim 10, Shalev-Shwartz, Vezhnevets, Kuefler, and Xu teach the method of claim 7, and Xu further teaches wherein the sequence of frames of image data is obtained from on-vehicle sensors (Xu: page 1, Section 1 (Introduction), "based on crowdsourced dashcam video sources"; page 5, Section 4.1, "The BDDV contains over 10,000 hours of driving dash-cam video streams"; the image frames are captured by a dashboard-mounted camera carried on the vehicle (on-vehicle sensors)).
Regarding dependent claim 17, the rejection of claim 7 is applied in the same manner to the corresponding system form recitation of claim 17.
Claims 8 and 9 are rejected under 35 U.S.C. 103 as being unpatentable over Shalev-Shwartz in view of Vezhnevets, Kuefler, and Xu, and further in view of Kundu et al. (hereinafter Kundu) “Realtime Multibody Visual SLAM with a Smoothly Moving Monocular Camera” (2011).
Regarding dependent claim 8, Shalev-Shwartz, in view of Vezhnevets, Kuefler, and Xu, teach the method of claim 7, together with the following portions of the recited reinforcement learning process.
Shalev-Shwartz teaches generating, from each of the one or more object trajectories, a plurality of trajectory sections (Shalev-Shwartz: page 4, Section 2, "denote a sub-trajectory from time stamp i to time stamp j"; a state-action sequence over a time period sufficient for long-term planning is divided at successive time stamps into sub-trajectories (a plurality of trajectory sections)), and Xu likewise teaches dividing recovered driving behaviour into successive short intervals over each of which an action is predicted (Xu: page 6, Section 5.1, "predicting the feasible actions in the next 1/3rd of a second").
Xu teaches determining a respective label for each trajectory section that represents one or more actions for the trajectory section (Xu: page 3, Section 3.1, "the motion action set A could be a set of coarse actions" (straight, stop, left-turn, right-turn); page 6, Section 5.1, "predicting the feasible actions in the next 1/3rd of a second"; each short interval of the trajectory is assigned the driving action carried out during it (a respective label ... that represents one or more actions for the trajectory section)).
Kuefler teaches generating a simulated observation in response to one or more actions of a simulated agent in a simulated environment (Kuefler: page 3, Section IV-B, "negative examples are sampled from rollouts generated by interactions of" the learned policy with the simulation environment; page 4, Section VI-A, "Simulations are initialized to match frames from the NGSIM data"; page 4, Section VI-A, "The ego vehicle is driven according to a bicycle model with acceleration and turn-rate sampled from the policy network"; the simulated ego vehicle (a simulated agent) selects an action from the policy and the simulation is advanced to yield the next simulated state in response to that action (a simulated observation), over successive steps in the driving simulation (a simulated environment)).
Vezhnevets teaches processing first data characterizing the simulated observation using the path planning network, according to current values of the first set of network parameters, to generate a first output specifying a planned trajectory for the simulated agent, and processing second data characterizing (i) the simulated observation and (ii) the planned trajectory generated by the path planning network using the high-level action control network, according to current values of the second set of network parameters, to generate a second output specifying one or more planned actions for the simulated agent (Vezhnevets: page 2, "The Manager internally computes a latent state representation" and outputs its goal; page 2, "conditioned on external observation, its own state, and the Managers goal"; page 3, "we ought to end up taking advantageous trajectories through state-space"; the Manager network (the path planning network) processes the simulated observation and outputs the planned trajectory, and the Worker network (the high-level action control network) produces the planned actions conditioned on the simulated observation and that planned trajectory, as set forth for claim 1).
Kuefler teaches determining a first similarity score characterizing a first similarity between (i) the planned trajectory generated by the path planning network and (ii) the object trajectories determined for the planned trajectory for the dynamic object in the real-world scene, and adjusting values of the first set of network parameters based on a first reward determined from the first similarity score (Kuefler: page 3, Section IV-B, "trained to discriminate between policy and expert state-action pairs"; page 3, Section IV-B, the discriminator reward can be used to drive the policy "into regions of the state-action space similar to those explored by" the expert; page 3, Section IV-B, "After performing rollouts with a given set of policy parameters" the "surrogate rewards" are calculated and "TRPO is used to perform a policy update"; the discriminator scores how similar the trajectories generated by the network are to the real-world expert trajectories (a first similarity score), and the surrogate reward determined from that score is used by the policy update to adjust the network parameters (adjusting values of the first set of network parameters based on a first reward determined from the first similarity score)).
Vezhnevets and Xu teach determining a second similarity score characterizing a second similarity between (i) the planned actions generated by the high-level action control network and (ii) the respective labels determined for the trajectory sections of the dynamic object in the real-world scene, and adjusting values of the second set of network parameters based on a second reward determined from the second similarity score (Vezhnevets: page 3, Section 3.2, the intrinsic reward is computed with the function that "is the cosine similarity between two vectors"; page 4, Section 3.2, the Worker policy "can be trained to maximise intrinsic reward by using any off-the shelf deep reinforcement learning algorithm"; Xu: page 3, Section 3.1, "the motion action set A could be a set of coarse actions" (straight, stop, left-turn, right-turn); the planned actions produced by the high-level action control network are scored by an explicit cosine-similarity measure (a second similarity score), that network is trained by an intrinsic reward determined from that similarity score (adjusting values of the second set of network parameters based on a second reward determined from the second similarity score), and in the combination the comparison target for that score is supplied by the real-world action labels that Xu determines for the trajectory sections (the respective labels determined for the trajectory sections of the dynamic object in the real-world scene)).
Shalev-Shwartz, Vezhnevets, Kuefler, and Xu do not expressly teach processing the image data to determine one or more object trajectories for a dynamic object in the real-world scene.
However, Kundu teaches processing the image data to determine one or more object trajectories for a dynamic object in the real-world scene (Kundu: page 2, Abstract, "a realtime, incremental multibody visual SLAM system that allows choosing between full 3D reconstruction or simply tracking of the moving objects"; page 3, Figure 1, "an integrated dynamic map of the scene including 3D structure and 3D trajectory of the camera, static world and moving objects"; the monocular image sequence of the real-world scene is processed by feature tracking and motion segmentation to determine the trajectories of the moving objects (processing the image data to determine one or more object trajectories for a dynamic object in the real-world scene)).
Because Shalev-Shwartz, in view of Vezhnevets, Kuefler, and Xu, and Kundu are analogous art in the same field of endeavor of recovering and modelling the motion of agents from real-world camera imagery, and each is reasonably pertinent to the problem of obtaining accurate object trajectories from monocular driving video, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to process the real-world driving video frames of Xu with the moving-object tracking of Kundu to determine the object trajectories of the dynamic objects in the scene, so that those recovered trajectories and their per-section action labels supply the two comparison targets of the similarity-based training process, with a reasonable expectation of success, to teach processing the image data to determine one or more object trajectories for a dynamic object in the real-world scene. This modification would have been motivated by the desire to supply objective real-world trajectory targets for the similarity-based training signals, as taught by Kundu (Kundu: page 2, "accounting for both the static and moving objects provides richer information about the environment").
Regarding dependent claim 9, Shalev-Shwartz, Vezhnevets, Kuefler, Xu, and Kundu teach the method of claim 8 as set forth above, and Kundu further teaches converting the one or more trajectories from two-dimensional space to three-dimensional space (Kundu: page 3, Figure 1, "an integrated dynamic map of the scene including 3D structure and 3D trajectory of the camera, static world and moving objects"; the two-dimensional monocular-image tracks of the moving objects are reconstructed into three-dimensional trajectories (converting the one or more trajectories from two-dimensional space to three-dimensional space)). The rejection relies on the same combination and the same motivations set forth for claim 8, the two-dimensional to three-dimensional conversion being taught by Kundu.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Bernal et al., US 2017/0169297 A1 (Jun. 15, 2017) (ABSTRACT A system and method of monitoring a region of interest comprises obtaining visual data comprising image frames of the region of interest over a period of time, analyzing individual subjects within the region of interest, the analyzing including at least one of tracking movement of individual subjects over time within the region of interest or extracting an appearance attribute of the individual subjects, and defining a group to include individual subjects having at least one of similar movement profiles or similar appearance attributes. The tracking movement includes detecting at least one of a trajectory of an individual subject within the region of interest, a dwell of an individual subject in at least one location within the region of interest, or an entrance or exit location within the region of interest).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to KUANG FU CHEN whose telephone number is (571)272-1393. The examiner can normally be reached M-F 9:00-5:30pm 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, Jennifer Welch can be reached on (571) 272-7212. 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.
/KC CHEN/Primary Patent Examiner, Art Unit 2143
Prosecution Insights