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 .
Response to Amendment
This action is in response to amendments and remarks filed on 07/01/2026. Claims 1-9, 11, and 13-19 are pending. Claims 10 and 12 have been cancelled. Claims 1, 5, 8-9, 11, 13-15, and 17-18 have been amended. The 35 U.S.C. 112 rejections to claims 5, 8, 11, and 14 have been withdrawn in light of the instant amendments. This action is made final, as necessitated by amendment.
Priority
Applicant' s claim for the benefit of a prior-filed application under 35 U.S.C. 119(e) or under 35 U.S.C. 120, 121, 365(c), or 386(c) is acknowledged.
Information Disclosure Statement
The information disclosure statement (IDS) filed on 04/24/2026 have been acknowledged.
Response to Arguments
Applicant presents the following main arguments regarding the previous office action:
Regarding claim 1, Applicant argues that Ré is silent regarding "a collection structure of time dependent neural network models each valid for an epoch corresponding to a portion of a trajectory" or "repeating the steps...to provide continuous neural network navigational control" as recited in amended claim 1. Therefore, Ré fails to teach claim 1. Applicant also argues that Ré fails to teach claim 2 because it is silent as to the "a collection structure of time dependent neural network models each valid for an epoch corresponding to a portion of a trajectory" as recited in amended claim 1. Applicant further argues claim 1’s dependent claims, claim 17, and claim 17’s dependent claims are not taught because Ré fails to teach claim 1.
Regarding claim 9, Applicant argues that Ré fails to teach claim 9 because Ré fails to teach “reading neural network model parameters from a second almanac, wherein the second almanac comprises a plurality of neural network models for contingency maneuvers in which the spacecraft has deviated substantially from a nominal trajectory, the contingency maneuvers being configured to return the spacecraft to the nominal trajectory" as is acknowledged in the previous Office Action at page 11.
Regarding the claims rejected using Berntorp and Santoni, Applicant argues that since Berntorp and Santoni are directed to navigational control for automated vehicle systems, one of skill in the art of spacecraft navigational control would not find it obvious to incorporate the teachings of Berntorp and Santoni due to the dynamic nature of automotive vehicle operation.
Regarding argument A, the argument has been fully considered but is not persuasive. Examiner agrees that Ré fails to explicitly recite every limitation in claim 1. However, Examiner believes that these deficiencies would have been obvious given Ré’s method. Applicant specifically states Ré fails to teach the limitations of "a collection structure of time dependent neural network models each valid for an epoch corresponding to a portion of a trajectory" or "repeating the steps...to provide continuous neural network navigational control".
However, Ré teaches "a collection structure of time dependent neural network models each valid for an epoch corresponding to a portion of a trajectory" (abstract, “a series of NNs, where each NN is applicable to a predefined range of states and/or epochs”—the NNs are applicable to states and/or epochs and are to be used on the trajectory of the spacecraft, therefore the NNs are time-dependent neural network models that are each valid for an epoch corresponding to a portion of a trajectory). While a specific collection structure is not disclosed, the NNs are stored on the spacecraft and would inherently have some sort of collection structure. The term “collection structure” is very broad and could include any number of ways the NNs are stored on the spacecraft.
Regarding the limitation "repeating the steps...to provide continuous neural network navigational control", as Applicant acknowledges in their arguments, Ré recites that "Trajectory corrections are made continuously over the course of an orbit transfer" (Ré pg. 9). Applicant argues that Ré teaches the corrections are made over one orbit transfer and not an entire spacecraft mission. Examiner agrees with Applicant’s conclusion. However, the claims do not require the corrections be done over an entire spacecraft mission, only that it provides “continuous neural network navigational control”. Therefore, Ré teaches this limitation. Further details can be found below under Claim Rejections.
Regarding argument B, the argument has been fully considered and is persuasive. However, upon further consideration, a new ground(s) of rejection is made in view of Ré, Berntorp, and Santoni. Further details can be found below under Claim Rejections.
Regarding argument C, the argument has been fully considered but is not persuasive. The claims do not preclude dynamic navigation. Therefore, there would be no reason for one of ordinary skill in the art to disregard references regarding automotive vehicle operation.
Claim Rejections - 35 USC § 103
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim(s) 1-2, 5, and 17-18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ré (Neural Networks for Onboard Maneuver Design).
Regarding claim 1, Ré teaches an almanac method for autonomous neural network navigation of a spacecraft, the method comprising:
providing an almanac (abstract, "Ground systems generate training data (consisting of tens of thousands of off nominal maneuver designs) and train a series of NNs, where each NN is applicable to a predefined range of states and/or epochs”) stored in memory of a computer onboard the spacecraft (abstract, "The framework’s computational burden for the spacecraft is minimal and easily fits within most current flight computers"), the almanac comprising:
1) a collection structure of time dependent neural network models each valid for an epoch corresponding to a portion of a trajectory (abstract, “a series of NNs, where each NN is applicable to a predefined range of states and/or epochs”),
2) model weights corresponding to each of the neural network models (abstract, “a series of NNs”—NNs would have model weights),
3) core logic configured to operate on the almanac for autonomous trajectory corrections (abstract, “a series of NNs, where each NN is applicable to a predefined range of states and/or epochs…The NN framework is also implemented in prototype flight software”—NNs are used to control the spacecraft), and
4) ancillary information configured to map each of the neural network models to a corresponding epoch range (abstract, “a series of NNs, where each NN is applicable to a predefined range of states and/or epochs…The NN framework is also implemented in prototype flight software”—NNs are used to control the spacecraft);
providing autonomous neural network navigation of the spacecraft (abstract, “neural network (NN) model to making a single low-thrust trajectory correction in cislunar space”) by performing steps via the computer onboard the spacecraft (abstract, “The framework’s computational burden for the spacecraft is minimal and easily fits within most current”), the steps comprising:
determining a current navigation state of the spacecraft (Fig. 13, Navigation update; pg. 9 right column, "The predicted spacecraft’s state is periodically updated to that of the truth");
determining a current target epoch for the spacecraft based on the current navigation state (abstract, “each NN is applicable to a predefined range of states and/or epochs”; pg. 9 right column, "At a defined NN evaluation frequency, the predicted spacecraft’s state is used to evaluate the appropriate NN and deliver the desired control");
identifying relevant model weights to read from the almanac (the NNs would have model weights);
reading neural network model parameters including relevant model weights from an almanac for the current target epoch (abstract, “a series of NNs, where each NN is applicable to a predefined range of states and/or epochs”);
and executing a neural network model of the plurality corresponding to the current target epoch to provide one or more thrust commands for the spacecraft (pg. 9 right column, "At a defined NN evaluation frequency, the predicted spacecraft’s state is used to evaluate the appropriate NN and deliver the desired control");
and repeating the steps for providing autonomous neural network navigation of the spacecraft for a next target epoch to provide continuous neural network navigational control (pg. 9 left column, “Trajectory corrections are made continuously over the course of an orbit transfer”).
Although not every element is explicitly taught in Ré, these deficiencies would have been obvious given Ré’s method. The use of a series of NNs located on the spacecraft would necessitate some form of collection structure for the NNs. The term “collection structure” is very broad and could include any number of ways the NNs are stored on the spacecraft. NNs use model weights, so such information would be stored and retrieved when the NNs are used. The NNs are obviously used for controlling the spacecraft, so there would be some form of logic used in order to determine the correct NN of the series of NNs. The NNs are applicable to a predefined range of states and/or epochs, therefore information regarding when to use which NN would need to be stored.
Regarding claim 2, Ré teaches the almanac method of claim 1. Ré further teaches the epoch corresponding to each of the neural network models overlaps with at least one neighboring epoch to provide continuous neural network navigational control throughout the trajectory (abstract, “each NN is applicable to a predefined range of states and/or epochs”; Fig. 13 and pg. 9 left column, “Trajectory corrections are made continuously over the course of an orbit transfer”).
Although Ré does not explicitly teach the epochs overlap, this would have been an obvious choice in order to provide the continuous control that Ré teaches.
Regarding claim 5, Ré teaches the almanac method of claim 1. Ré further teaches the collection structure of time dependent neural network models comprises at least one pair of time-based neural network models configured for use with paired trajectory correction maneuvers (pg. 5 right column, “GEO station keeping using chemical propulsion consists of two separate maneuver types: an east-west maneuver that corrects longitudinal drift and a north-south maneuver that corrects latitudinal drift”; pg. 6 left column, “A separate NN is trained for each maneuver type”; pg. 8, “A separate feedforward NN is used to learn each of the five nominal TCMs”—although this example has 5 models for 5 TCMs, this would obviously be applicable to a mission with a different number of TCMs).
Regarding claim 17, Ré teaches a computer program product for autonomous neural network navigation of a spacecraft (abstract, “neural network (NN) model to making a single low-thrust trajectory correction in cislunar space”), the computer program product comprising a computer readable storage medium having computer readable instructions stored therein (abstract, “The framework’s computational burden for the spacecraft is minimal and easily fits within most current”), wherein the computer readable instructions, when executed on a computing device, cause the computing device to:
read neural network model parameters from an almanac stored in the computer readable storage medium (abstract, “a series of NNs, where each NN is applicable to a predefined range of states and/or epochs…The NN framework is also implemented in prototype flight software”), the almanac comprising:
1) a collection structure of time dependent neural network models, wherein each neural network model is valid for an epoch corresponding to a portion of a trajectory (abstract, “a series of NNs, where each NN is applicable to a predefined range of states and/or epochs”);
2) model weights corresponding to each of the neural network models (abstract, “a series of NNs”—NNs would have model weights),
3) core logic configured to operate on the almanac for autonomous trajectory corrections (abstract, “a series of NNs, where each NN is applicable to a predefined range of states and/or epochs…The NN framework is also implemented in prototype flight software”—NNs are used to control the spacecraft), and
4) ancillary information configured to map each of the neural network models to a corresponding epoch range (abstract, “a series of NNs, where each NN is applicable to a predefined range of states and/or epochs…The NN framework is also implemented in prototype flight software”—NNs are used to control the spacecraft);
determine a current navigation state of the spacecraft (Fig. 13, Navigation update; pg. 9 right column, "The predicted spacecraft’s state is periodically updated to that of the truth");
determine a current target epoch for the spacecraft based on the current navigation state (abstract, “each NN is applicable to a predefined range of states and/or epochs”; pg. 9 right column, "At a defined NN evaluation frequency, the predicted spacecraft’s state is used to evaluate the appropriate NN and deliver the desired control");
identify relevant model weights to read from the almanac (the NNs would have model weights);
read neural network model parameters including relevant model weights from an almanac for the current target epoch (abstract, “a series of NNs, where each NN is applicable to a predefined range of states and/or epochs”);
and execute the neural network model corresponding to the current target epoch to provide one or more thrust commands for the spacecraft (pg. 9 right column, "At a defined NN evaluation frequency, the predicted spacecraft’s state is used to evaluate the appropriate NN and deliver the desired control").
Although not every element is explicitly taught in Ré, these deficiencies would have been obvious given Ré’s method. The use of a series of NNs located on the spacecraft would necessitate some form of collection structure for the NNs. The term “collection structure” is very broad and could include any number of ways the NNs are stored on the spacecraft. NNs use model weights, so such information would be stored and retrieved when the NNs are used. The NNs are obviously used for controlling the spacecraft, so there would be some form of logic used in order to determine the correct NN of the series of NNs. The NNs are applicable to a predefined range of states and/or epochs, therefore information regarding when to use which NN would need to be stored.
Regarding claim 18, Ré teaches the computer program product of claim 17. Ré further teaches the collection structure of time dependent neural network models is configured to provide continuous neural network navigational control throughout the entire trajectory (abstract, “each NN is applicable to a predefined range of states and/or epochs”; Fig. 13 and pg. 9 left column, “Trajectory corrections are made continuously over the course of an orbit transfer”).
Claim(s) 3 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ré in view of Gu (Optimal orbit transfer of spacecraft under constant thrust).
Regarding claim 3, Ré teaches the almanac method of claim 2. Ré fails to explicitly teach the neural network navigational control comprises continuous thrust across a plurality of epochs with no coasting.
However, using constant thrust throughout a trajectory is already well-known in the field and would have been an obvious design choice. For example, Gu teaches continuous thrust across a plurality of epochs with no coasting (abstract, “This paper considers the optimal orbit transfer problem of spacecraft under constant thrust”).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Ré to incorporate the teachings of Gu, which states, “The main purpose is to design the optimal control law to ensure the orbit transfer of the spacecraft with the least time consumption” (abstract).
Claim(s) 4 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ré in view of Parrish (Low Thrust Trajectory Optimization in Cislunar and Translunar Space)
Regarding claim 4, Ré teaches the almanac method of claim 1. Ré fails to explicitly teach the epochs corresponding to the neural network models do not overlap such that no thrust commands are provided between epochs and the spacecraft coasts between epochs.
However, coasting between thrusts is already well-known in the field and would have been an obvious design choice. For example, Parrish teaches no thrust commands are provided between epochs and the spacecraft coasts between epochs (pg. 40, "bang-coast-bang").
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Ré to incorporate the teachings of Parrish since a ban-coast-bang thrust structure would be fuel-optimal (pg. 40).
Claim(s) 6-8, 9, 11, 13-16, and 19-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Ré in view of Berntorp (US 20180120843 A1) and Santoni (US 20200017114 A1).
Regarding claim 6, Ré teaches the almanac method of claim 1. Ré fails to teach reading neural network model parameters from a second almanac, wherein the second almanac comprises a plurality of neural network models for contingency maneuvers in which the spacecraft has deviated substantially from a nominal trajectory, the contingency maneuvers being configured to return the spacecraft to the nominal trajectory.
However, Berntorp teaches reading neural network model parameters (par. 45, “the system 99 can select the neural network based on the time-series signals 131 itself. For example, for different driving situations, for example, given by an external input 110, different neural networks are selected 141; par. 49, “The method selects 175, e.g., from the memory 140, a neural network 172 trained to transform time-series signals to reference trajectories of the vehicle and determines 175 the reference trajectory 176 submitting the time-series signal to the neural network. The neural network is trained to produce the reference trajectory 176 as a function of time that satisfies time and spatial constraints on a position of the vehicle…Examples of the time and spatial constraints include a bound on a deviation of a location of the vehicle from a middle of a road, a bound of deviations from a desired location at a given time step, a bound on a deviation of the time when reaching a desired location, a minimal distance to an obstacle on the road, and the time it should take to complete a lane change”).
Berntorp teaches selecting from a plurality of neural network models (par. 44, “the memory 140 stores a set of neural networks, each neural network in the set is trained to consider different driving styles for mapping the time-series signals to reference trajectories of the vehicle”) which can be based on a deviation from a desired location on a trajectory (par. 45, “Additionally, or alternatively, the system 99 can select the neural network based on the time-series signals 131 itself”). Both Ré and Berntorp are directed to a method for controlling the trajectory of a vehicle. While Ré is specifically directed to a spacecraft, methods directed towards other types of vehicles would obviously be relevant. It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Ré to incorporate the teachings of Berntorp in order to “optimize some criteria associated to the operation of the vehicle” (par. 2) and to “streamline the process for determining and controlling the motion of the vehicle” (par. 4).
Both Ré and Berntorp fail to explicitly teach a second almanac, wherein the second almanac comprises a plurality of neural network models for contingency maneuvers. However, using a separate method or system for solving deviation errors is already well-known in the field.
Santoni teaches an automatic driving system for a vehicle that uses a separate system for determining contingency maneuvers (par. 51, “the safety companion 710 (e.g., in response to detecting a fatal error or multiple or repeated errors by the compute subsystem's processing complex over a time period) may engage a failover automated driving system (e.g., 750)”).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Ré in view of Berntorp to incorporate the teachings of Santoni to have Berntorp’s contingency neural network models be from a separate almanac. Santoni discloses “a safety monitor application 810 may be provided on the safety companion subsystem 710 to implement logic (e.g., executable by safety companion processing hardware (e.g., 720a, 720b)) to…execute simplified automated driving operations (and/or invoke internal failover control logic (e.g., 750) or a more robust failover automated driving system provided on the vehicle) in an attempt to remedy or mitigate effects of errors or other issues determined to affect the safety of the automated driving decisions driven by the compute subsystem 705” (par. 53). It would obviously be beneficial to have a more robust system for larger errors.
Regarding claim 7, the combination of Ré in view of Berntorp and Santoni teaches the almanac method of claim 6. Ré fails to teach when the spacecraft has deviated substantially from the nominal trajectory, an appropriate contingency model is looked up based on an amount of deviation of the current navigation state from an expected nominal state.
However, Berntorp teaches when the spacecraft has deviated substantially from the nominal trajectory, an appropriate contingency model is looked up based on an amount of deviation of the current navigation state from an expected nominal state (par. 45, “the system 99 can select the neural network based on the time-series signals 131 itself. For example, for different driving situations, for example, given by an external input 110, different neural networks are selected 141; par. 49, “The method selects 175, e.g., from the memory 140, a neural network 172 trained to transform time-series signals to reference trajectories of the vehicle and determines 175 the reference trajectory 176 submitting the time-series signal to the neural network. The neural network is trained to produce the reference trajectory 176 as a function of time that satisfies time and spatial constraints on a position of the vehicle…Examples of the time and spatial constraints include a bound on a deviation of a location of the vehicle from a middle of a road, a bound of deviations from a desired location at a given time step, a bound on a deviation of the time when reaching a desired location, a minimal distance to an obstacle on the road, and the time it should take to complete a lane change”).
Berntorp teaches selecting from a plurality of neural network models (par. 44, “the memory 140 stores a set of neural networks, each neural network in the set is trained to consider different driving styles for mapping the time-series signals to reference trajectories of the vehicle”) which can be based on a deviation from a desired location on a trajectory (par. 45, “Additionally, or alternatively, the system 99 can select the neural network based on the time-series signals 131 itself”; par. 49, “Examples of the time and spatial constraints include…a bound of deviations from a desired location at a given time step”). It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Ré in view of Berntorp and Santoni to further incorporate the teachings of Berntorp in order to “optimize some criteria associated to the operation of the vehicle” (par. 2) and to “streamline the process for determining and controlling the motion of the vehicle” (par. 4).
Regarding claim 8, the combination of Ré in view of Berntorp and Santoni teaches the almanac method of claim 6. Both Ré and Berntorp fail to explicitly teach the neural network models for contingency maneuvers are more robust to large errors but less accurate compared to nominal neural network models.
However, Santoni teaches the neural network models for contingency maneuvers are more robust to large errors but less accurate compared to nominal neural network models (par. 53, “a safety monitor application 810 may be provided on the safety companion subsystem 710 to implement logic (e.g., executable by safety companion processing hardware (e.g., 720a, 720b)) to…execute simplified automated driving operations (and/or invoke internal failover control logic (e.g., 750) or a more robust failover automated driving system provided on the vehicle) in an attempt to remedy or mitigate effects of errors or other issues determined to affect the safety of the automated driving decisions driven by the compute subsystem 705”) .
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Ré in view of Berntorp and Santoni to further incorporate the teachings of Santoni. It would obviously be beneficial to have a more robust system for larger errors.
Regarding claim 9, Ré teaches a method for autonomous neural network navigation of a spacecraft, the method comprising:
providing a first almanac (abstract, "Ground systems generate training data (consisting of tens of thousands of off nominal maneuver designs) and train a series of NNs, where each NN is applicable to a predefined range of states and/or epochs”) stored in memory of a computer onboard the spacecraft (abstract, "The framework’s computational burden for the spacecraft is minimal and easily fits within most current flight computers"), wherein the first almanac comprises a first plurality of time-based neural network models configured for nominal maneuvers in which the spacecraft remains within a nominal trajectory (abstract, “a series of NNs, where each NN is applicable to a predefined range of states and/or epochs”),
performing autonomous navigational steps (abstract, “neural network (NN) model to making a single low-thrust trajectory correction in cislunar space”) via the computer onboard the spacecraft (abstract, “The framework’s computational burden for the spacecraft is minimal and easily fits within most current”), the steps comprising:
determining a current navigation state of the spacecraft (Fig. 13, Navigation update; pg. 9 right column, "The predicted spacecraft’s state is periodically updated to that of the truth");
reading neural network model parameters including relevant model weights from a) the first almanac for the current target epoch when the spacecraft is within the nominal trajectory (abstract, “a series of NNs, where each NN is applicable to a predefined range of states and/or epochs”),
and executing the neural network model corresponding to the current target epoch from the first (pg. 9 right column, "At a defined NN evaluation frequency, the predicted spacecraft’s state is used to evaluate the appropriate NN and deliver the desired control").
Although not every element is explicitly taught in Ré, these deficiencies would have been obvious given Ré’s method. The use of a series of NNs located on the spacecraft would necessitate some form of collection structure for the NNs. The term “collection structure” is very broad and could include any number of ways the NNs are stored on the spacecraft. NNs use model weights, so such information would be stored and retrieved when the NNs are used. The NNs are obviously used for controlling the spacecraft, so there would be some form of logic used in order to determine the correct NN of the series of NNs. The NNs are applicable to a predefined range of states and/or epochs, therefore information regarding when to use which NN would need to be stored.
Ré fails to teach providing a second almanac stored in memory of the computer onboard the spacecraft, wherein the second almanac comprises a second plurality of time-based neural network models configured for contingency maneuvers to return the spacecraft to the nominal trajectory; determining whether the spacecraft is within the nominal trajectory; reading neural network model parameters including relevant model weights from the second almanac for the current target epoch when the spacecraft is outside the nominal trajectory; and executing the neural network model corresponding to the current target epoch from the second almanac.
However, Berntorp teaches providing a (par. 45, “the system 99 can select the neural network based on the time-series signals 131 itself. For example, for different driving situations, for example, given by an external input 110, different neural networks are selected 141; par. 49, “The method selects 175, e.g., from the memory 140, a neural network 172 trained to transform time-series signals to reference trajectories of the vehicle and determines 175 the reference trajectory 176 submitting the time-series signal to the neural network. The neural network is trained to produce the reference trajectory 176 as a function of time that satisfies time and spatial constraints on a position of the vehicle…Examples of the time and spatial constraints include a bound on a deviation of a location of the vehicle from a middle of a road, a bound of deviations from a desired location at a given time step, a bound on a deviation of the time when reaching a desired location, a minimal distance to an obstacle on the road, and the time it should take to complete a lane change”);
determining whether the spacecraft is within the nominal trajectory (par. 45, “Additionally, or alternatively, the system 99 can select the neural network based on the time-series signals 131 itself”);
reading neural network model parameters including relevant model weights from the (par. 45, “Additionally, or alternatively, the system 99 can select the neural network based on the time-series signals 131 itself”—NN would have model parameters and model weights);
and executing the neural network model corresponding to the current target epoch from the (Fig. 1E step 190, controlling the motion of the vehicle).
Berntorp teaches selecting from a plurality of neural network models (par. 44, “the memory 140 stores a set of neural networks, each neural network in the set is trained to consider different driving styles for mapping the time-series signals to reference trajectories of the vehicle”) which can be based on a deviation from a desired location on a trajectory (par. 45, “Additionally, or alternatively, the system 99 can select the neural network based on the time-series signals 131 itself”). Both Ré and Berntorp are directed to a method for controlling the trajectory of a vehicle. While Ré is specifically directed to a spacecraft, methods directed towards other types of vehicles would obviously be relevant. It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Ré to incorporate the teachings of Berntorp in order to “optimize some criteria associated to the operation of the vehicle” (par. 2) and to “streamline the process for determining and controlling the motion of the vehicle” (par. 4).
Both Ré and Berntorp fail to explicitly teach a second almanac, wherein the second almanac comprises a plurality of neural network models for contingency maneuvers. However, using a separate method or system for solving deviation errors is already well-known in the field.
Santoni teaches an automatic driving system for a vehicle that uses a separate system for determining contingency maneuvers (par. 51, “the safety companion 710 (e.g., in response to detecting a fatal error or multiple or repeated errors by the compute subsystem's processing complex over a time period) may engage a failover automated driving system (e.g., 750)”).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Ré in view of Berntorp to incorporate the teachings of Santoni to have Berntorp’s contingency neural network models be from a separate almanac. Santoni discloses “a safety monitor application 810 may be provided on the safety companion subsystem 710 to implement logic (e.g., executable by safety companion processing hardware (e.g., 720a, 720b)) to…execute simplified automated driving operations (and/or invoke internal failover control logic (e.g., 750) or a more robust failover automated driving system provided on the vehicle) in an attempt to remedy or mitigate effects of errors or other issues determined to affect the safety of the automated driving decisions driven by the compute subsystem 705” (par. 53). It would obviously be beneficial to have a more robust system for larger errors.
Regarding claim 11, the combination of Ré in view of Berntorp and Santoni teaches the method of claim 9. Ré further teaches the first plurality of time-based neural network models comprises a pair of time-based neural network models configured for use with paired trajectory correction maneuvers (pg. 5 right column, “GEO station keeping using chemical propulsion consists of two separate maneuver types: an east-west maneuver that corrects longitudinal drift and a north-south maneuver that corrects latitudinal drift”; pg. 6 left column, “A separate NN is trained for each maneuver type”; pg. 8, “A separate feedforward NN is used to learn each of the five nominal TCMs”—although this example has 5 models for 5 TCMs, this would obviously be applicable to a mission with a different number of TCMs).
Regarding claim 13, the combination of Ré in view of Berntorp and Santoni teaches the method of claim 9. Ré fails to teach when the spacecraft has deviated from the nominal trajectory, an appropriate contingency model is looked up based on an amount of deviation of the current navigation state from an expected nominal state.
However, Berntorp teaches when the spacecraft has deviated from the nominal trajectory, an appropriate contingency model is looked up based on an amount of deviation of the current navigation state from an expected nominal state (par. 45, “the system 99 can select the neural network based on the time-series signals 131 itself. For example, for different driving situations, for example, given by an external input 110, different neural networks are selected 141; par. 49, “The method selects 175, e.g., from the memory 140, a neural network 172 trained to transform time-series signals to reference trajectories of the vehicle and determines 175 the reference trajectory 176 submitting the time-series signal to the neural network. The neural network is trained to produce the reference trajectory 176 as a function of time that satisfies time and spatial constraints on a position of the vehicle…Examples of the time and spatial constraints include a bound on a deviation of a location of the vehicle from a middle of a road, a bound of deviations from a desired location at a given time step, a bound on a deviation of the time when reaching a desired location, a minimal distance to an obstacle on the road, and the time it should take to complete a lane change”).
Berntorp teaches selecting from a plurality of neural network models (par. 44, “the memory 140 stores a set of neural networks, each neural network in the set is trained to consider different driving styles for mapping the time-series signals to reference trajectories of the vehicle”) which can be based on a deviation from a desired location on a trajectory (par. 45, “Additionally, or alternatively, the system 99 can select the neural network based on the time-series signals 131 itself”; par. 49, “Examples of the time and spatial constraints include…a bound of deviations from a desired location at a given time step”). It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Ré in view of Berntorp and Santoni to further incorporate the teachings of Berntorp in order to “optimize some criteria associated to the operation of the vehicle” (par. 2) and to “streamline the process for determining and controlling the motion of the vehicle” (par. 4).
Regarding claim 14, the combination of Ré in view of Berntorp and Santoni teaches the method of claim 13. Both Ré and Berntorp fail to explicitly teach the appropriate contingency model is more robust to deviations of the current navigation state from the expected nominal state, but less accurate compared to neural network models used for the nominal trajectory.
However, Santoni teaches the appropriate contingency model is more robust to deviations of the current navigation state from the expected nominal state, but less accurate compared to neural network models used for the nominal trajectory (par. 53, “a safety monitor application 810 may be provided on the safety companion subsystem 710 to implement logic (e.g., executable by safety companion processing hardware (e.g., 720a, 720b)) to…execute simplified automated driving operations (and/or invoke internal failover control logic (e.g., 750) or a more robust failover automated driving system provided on the vehicle) in an attempt to remedy or mitigate effects of errors or other issues determined to affect the safety of the automated driving decisions driven by the compute subsystem 705”).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Ré in view of Berntorp and Santoni to further incorporate the teachings of Santoni. It would obviously be beneficial to have a more robust system for larger errors.
Regarding claim 15, the combination of Ré in view of Berntorp and Santoni teaches the method of claim 9. Ré further teaches the first almanac comprises a single neural network model configured to provide autonomous spacecraft navigation from a mission start to a mission end (pg. 7 right column, “A single feedforward neural network is trained for all OMM epochs”).
Regarding claim 16, the combination of Ré in view of Berntorp and Santoni teaches the method of claim 15. Ré further teaches the single neural network model comprises one or more thrust commands configured for providing one or more spacecraft maneuvers (pg. 7 left column, “perform an orbital maintenance maneuver (OMM)”).
Regarding claim 19, Ré teaches the computer program product of claim 17. Ré fails to teach the computer readable instructions are configured to read neural network model parameters from a second almanac, wherein the second almanac comprises a plurality of neural network models for contingency maneuvers in which the spacecraft has deviated substantially from a nominal trajectory, the contingency maneuvers being configured to return the spacecraft to the nominal trajectory.
However, Berntorp teaches the computer readable instructions are configured to read neural network model parameters (par. 45, “the system 99 can select the neural network based on the time-series signals 131 itself. For example, for different driving situations, for example, given by an external input 110, different neural networks are selected 141; par. 49, “The method selects 175, e.g., from the memory 140, a neural network 172 trained to transform time-series signals to reference trajectories of the vehicle and determines 175 the reference trajectory 176 submitting the time-series signal to the neural network. The neural network is trained to produce the reference trajectory 176 as a function of time that satisfies time and spatial constraints on a position of the vehicle…Examples of the time and spatial constraints include a bound on a deviation of a location of the vehicle from a middle of a road, a bound of deviations from a desired location at a given time step, a bound on a deviation of the time when reaching a desired location, a minimal distance to an obstacle on the road, and the time it should take to complete a lane change”).
Berntorp teaches selecting from a plurality of neural network models (par. 44, “the memory 140 stores a set of neural networks, each neural network in the set is trained to consider different driving styles for mapping the time-series signals to reference trajectories of the vehicle”) which can be based on a deviation from a desired location on a trajectory (par. 45, “Additionally, or alternatively, the system 99 can select the neural network based on the time-series signals 131 itself”). It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Ré to incorporate the teachings of Berntorp in order to “optimize some criteria associated to the operation of the vehicle” (par. 2) and to “streamline the process for determining and controlling the motion of the vehicle” (par. 4).
Both Ré and Berntorp fail to explicitly teach a second almanac, wherein the second almanac comprises a plurality of neural network models for contingency maneuvers. However, using a separate method or system for solving deviation errors is already well-known in the field.
Santoni teaches an automatic driving system for a vehicle that uses a separate system for determining contingency maneuvers (par. 51, “the safety companion 710 (e.g., in response to detecting a fatal error or multiple or repeated errors by the compute subsystem's processing complex over a time period) may engage a failover automated driving system (e.g., 750)”).
It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Ré in view of Berntorp to incorporate the teachings of Santoni to have Berntorp’s contingency neural network models be from a separate almanac. Santoni discloses “a safety monitor application 810 may be provided on the safety companion subsystem 710 to implement logic (e.g., executable by safety companion processing hardware (e.g., 720a, 720b)) to…execute simplified automated driving operations (and/or invoke internal failover control logic (e.g., 750) or a more robust failover automated driving system provided on the vehicle) in an attempt to remedy or mitigate effects of errors or other issues determined to affect the safety of the automated driving decisions driven by the compute subsystem 705” (par. 53). It would obviously be beneficial to have a more robust system for larger errors.
Regarding claim 20, the combination of Ré in view of Berntorp and Santoni teaches the computer program product of claim 19. Ré fails to teach when the spacecraft has deviated substantially from the nominal trajectory, an appropriate contingency model is looked up based on an amount of deviation of the current navigation state from an expected nominal state.
However, Berntorp teaches when the spacecraft has deviated substantially from the nominal trajectory, an appropriate contingency model is looked up based on an amount of deviation of the current navigation state from an expected nominal state (par. 45, “the system 99 can select the neural network based on the time-series signals 131 itself. For example, for different driving situations, for example, given by an external input 110, different neural networks are selected 141; par. 49, “The method selects 175, e.g., from the memory 140, a neural network 172 trained to transform time-series signals to reference trajectories of the vehicle and determines 175 the reference trajectory 176 submitting the time-series signal to the neural network. The neural network is trained to produce the reference trajectory 176 as a function of time that satisfies time and spatial constraints on a position of the vehicle…Examples of the time and spatial constraints include a bound on a deviation of a location of the vehicle from a middle of a road, a bound of deviations from a desired location at a given time step, a bound on a deviation of the time when reaching a desired location, a minimal distance to an obstacle on the road, and the time it should take to complete a lane change”).
Berntorp teaches selecting from a plurality of neural network models (par. 44, “the memory 140 stores a set of neural networks, each neural network in the set is trained to consider different driving styles for mapping the time-series signals to reference trajectories of the vehicle”) which can be based on a deviation from a desired location on a trajectory (par. 45, “Additionally, or alternatively, the system 99 can select the neural network based on the time-series signals 131 itself”; par. 49, “Examples of the time and spatial constraints include…a bound of deviations from a desired location at a given time step”). It would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Ré in view of Berntorp and Santoni to further incorporate the teachings of Berntorp in order to “optimize some criteria associated to the operation of the vehicle” (par. 2) and to “streamline the process for determining and controlling the motion of the vehicle” (par. 4).
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MINATO LEE HORNER whose telephone number is (571)272-5425. The examiner can normally be reached M-F 8-5.
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, Christian Chace can be reached at (571) 272-4190. 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.
/M.L.H./Examiner, Art Unit 3665 /CHRISTIAN CHACE/Supervisory Patent Examiner, Art Unit 3665
Prosecution Insights