DETAILED ACTION
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 Remarks
Claim Rejections - 35 USC § 103
Applicant's arguments filed 02/18/2026 have been fully considered but they are not persuasive.
First, applicant's arguments appear to be against the references individually. One cannot show nonobviousness by attacking references individually where the rejections are based on combinations of references. See In re Keller, 642 F.2d 413, 208 USPQ 871 (CCPA 1981); In re Merck & Co., 800 F.2d 1091, 231 USPQ 375 (Fed. Cir. 1986).
The rejection provided in the previous Office Action dated 11/19/2025 made clear what the combination contemplated between Hoshyari and Hayashi was. To repeat what was found in 17. Of the previous Office Action:
“It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to train and utilize a machine learning model such as that taught by Hayashi to perform the optimization taught by Hoshyari without the need to perform the optimization method each and every time to allow for faster or even on-line/real-time processing from a target input motion to a final retargeted motion, or alternatively and/or as part of such process, to utilize the optimization model of Hoshyari within the training of a machine learning model such as in Hayashi with a reasonable expectation of success” (emphasis added)
Applicant’s arguments appear to be directed towards Hoshyari alone, and fail to make clear how Hoshyari does not teach the limitations indicated, even when at least some of the limitations recite features which were addressed in some manner in the previous Office Action. Hoshyari is relied upon for the general framework of vibration suppression through optimization of control inputs/parameters to be provided to a real robot via optimization using a differentiable simulator to accurately replicate the anticipated behavior of the real-world robot. In combination with Hayashi, which focuses on optimization using a machine learning model, the combination is of leveraging a machine learning model (which for Hayashi is vibration suppression specific) to replicate the optimization function of Hoshyari. The use of the model, including and especially data produced thereby, to be replicated under machine learning is routine. It is difficult to replicate the results of a model, particularly a mathematical model, without the model on hand and knowing what it inputs and outputs. Additionally, there is no expectation that the basic design concepts of optimization change in going from the mathematical model to the machine learned model replicating it; e.g. the desired output remains a motion most closely replicating the target animation.
Second, Applicant’s arguments appear to rely on features not present within the claims.
Applicant recites in the Remarks, “The cited art fails to disclose these features” referring to the new amended limitations and then states in reference thereto, “For example, Hoshyari uses simulation to test whether human-provided weights for vibration suppression were helpful. On the contrary, the pending application uses a prediction method in the simulation to predict where vibration will occur and then update the model, requiring less human input compared to Hoshyari”. There is no recitation of a “prediction method” within the new “simulating” limitations, or even in any of the claims. If this refers to the machine learning model, see again above wherein the combination of Hoshyari with Hayashi is pertinent and arguments to Hoshyari alone are insufficient.
Third, Applicant then states “Unlike claim 8, Hoshyari does not disclose providing unfiltered control parameters for the robotic system to the differentiable simulator to generate states of the robotic system over time, with the resulting states compared against an input animation and optimized relative to vibration suppression and the input animation to generate filtered control parameters that are vibration suppressed”. This appears wholly conclusory, with no showing of how the disclosure of Hoshyari does not disclose these features. These features are clearly disclosed by Hoshyari. See Section 6. of Hoshyari in general, portions of which have been recited in previous Office Actions. A current set of control parameters are simulated, see Parameterization subsection, in particular the statement “We initialize the control points by fitting the parameterized profiles to the input animation”. These are control parameters. Applicant does not appear to clearly disclose what is meant by “unfiltered”. Reviewing the disclosure, the best understanding of the distinction between “unfiltered” and “filtered” items including “control parameters” is that items are first “unfiltered” and when processed, optimized, altered, adjusted, or similarly changed are “filtered”. Thus, it is clear that Hoshyari discloses “unfiltered control parameters”. Hoshyari was previously shown to disclose a differential/differentiable simulator with respect to Claim 21 which Applicant does not respond to. The states of the input animation and the simulation are furthermore clearly compared, and based on the comparison optimized. See at least Regularization subsection, in particular the statement, “Our objective measures performance w.r.t. absolute coordinates. To provide a means to penalize relative differences, i.e., closeness to the artistic input, we formulate a regularizer that compares the current profile to the input profiles”. Thus, optimized control parameters are generated (which are consequently “filtered”) which suppress vibration (and thus “vibration suppressed”) based on comparing with the input animation.
Fourth, the present construction of the claim limitations suggests, and furthermore appears argued by Applicant (see “prediction method in the simulation”), that all of the newly amended steps/features are part of the simulating. Applicant’s cited support, which Examiner agrees appears to be the closest support for these amended features, directly contradicts these limitations as being supported in the manner claimed, specifically that all are of “the simulating”. See the related 112(a) rejection below.
Fifth, the particular combination of Hoshyari with Hayashi or any other reference with respect to these limitations appears particularly and especially open to broad interpretation. The claim presently claims “to create simulated vibration data” which 1) may be construed as the intended purpose to the preceding limitation, and 2) the following limitations of “the simulating comprises” never recites “the simulated vibration data” in any manner. Consequently, only one or even none of the steps may create or be related to the creation of “the simulated vibration data” (see again use of “comprising”). In other words, the steps following “comprising” are not integrated into the rest of the claim in a manner which provides a particularly narrow scope. Instead, they presently articulate steps which may have no relation to the other claim limitations. In summary, a combination of references may similarly provide un-linked or un-combined features wherein the reason for doing so may be simple redundancy or similar.
Claim Rejections - 35 USC § 112(a)
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention.
Claims 8, 10 – 18, and 20 – 22 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for applications subject to pre-AIA 35 U.S.C. 112, the inventor(s), at the time the application was filed, had possession of the claimed invention.
Claim 8 recites the limitations:
“wherein the simulating comprises:
providing unfiltered control parameters for the robotic system to a differentiable simulator to generate states of the robotic system over time;
comparing the states against an input animation; and
optimizing the states relative to vibration suppression and the input animation to generate filtered control parameters that are vibration suppressed”
Per [0147] – [0148] which are cited by Applicant as the support for these limitations and appear to the Examiner to be the closest disclosure, only the first of the three steps/features claimed appears to be of “simulating”, and the furthermore and separately the optimization step appears to be recited in a manner different from disclosed. Mapping the limitations to the reference characters provided in [0147] – [0148] and found in Figure 15 which is referred to therein, the limitations appear to map to the disclosure as follows:
“wherein the simulating comprises:
providing unfiltered control parameters for the robotic system (1510) to a differentiable simulator (1520) to generate states of the robotic system over time (1524);
comparing the states (1524) against an input animation (1534) (1540 limitation overall); and
optimizing (1550) the states (not found/supported) relative to vibration suppression and the input animation to generate filtered control parameters that are vibration suppressed (1570)”.
Based on the mapping, therefore, only the first limitation/step of “providing … over time” appears to be a part of simulation. The comparison and optimization steps are separate from the simulation as illustrated in Figure 15.
Furthermore, and separately, the optimization is not recited as being of “the states relative to vibration suppression and the input animation”. Instead, [0147] recites “The optimization step 1550 takes the output of comparison step 1540 to modify the unfiltered or input control parameters 1560 to generate filtered control parameters or output 1570 with vibration suppression control”. In other words, the control parameters are what is optimized based on the output of the comparison.
Examiner finally notes that wherein the limitations are not integrated into a clear whole (no recitation of simulated vibration data” provided, the overarching “simulating comprises” is interpreted in the interest of compact prosecution under prior art below as having limited narrowing scope.
Therefore, the claim contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for pre-AIA the inventor(s), at the time the application was filed, had possession of the claimed invention.
Regarding Claims 10 – 18, and 20 – 22, the claims depend from claim(s) rejected above and inherit the deficiencies of said claim(s) as described above. Therefore, Claims 10 – 18, and 20 – 22 are rejected under the same logic presented above.
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.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 8, 10 – 18, and 20 – 21 are rejected under 35 U.S.C. 103 as being unpatentable over Hoshyari et al. (Hoshyari, Shayan, et al. "Vibration-minimizing motion retargeting for robotic characters." ACM Transactions on Graphics (TOG) 38.4 (2019): 1-14.) in view of Hayashi et al. (US 20200338724 A1).
Regarding Claim 8, Hoshyari teaches:
A method of controlling a robotic system comprising:
simulating movement and vibration of the robotic system to create simulated vibration data (See at least Page 7 of Hoshyari, “To retarget artist-specified input onto our physical robots, we seek to minimize differences between simulated and target states, putting a priority on the suppression of visible vibrations of large amplitude. To do so, we parameterize and optimize the time-varying motor angles, and solve for their optimal control”), wherein the simulating comprises:
providing unfiltered control parameters for the robotic system to a differentiable simulator to generate states of the robotic system over time (See at least Page 7 of Hoshyari, “We initialize the control points by fitting the parameterized profiles to the input animation, collecting the spline parameters of all motors in a global parameter vector p”);
comparing the states against an input animation (Part of the optimization process, see below/next); and
optimizing the states relative to vibration suppression and the input animation to generate filtered control parameters that are vibration suppressed (See at least Page 7 of Hoshyari “To retarget artist-specified input onto our physical robots, we seek to minimize differences between simulated and target states, putting a priority on the suppression of visible vibrations of large amplitude” and “Our objective measures performance w.r.t. absolute coordinates. To provide a means to penalize relative differences, i.e., closeness to the artistic input, we formulate a regularizer that compares the current profile to the input profiles”)
…
generating a first set of control signals corresponding to an initial motion output based on an animation sequence to be performed by the robotic system (See at least Section 6, Page 7, “We represent the time-varying angle θi(t, pi) of motor i (hereafter referred to as i’s motor profile) with a spline interpolation, parameterized with control points pi … We initialize the control points by fitting the parameterized profiles to the input animation, collecting the spline parameters of all motors in a global parameter vector p” and caption of Figure 1 on Page 1 “We present a method for retargeting fast and dynamic animations onto physical robot characters, where the motor trajectories are optimized in order to suppress unwanted structural vibrations and match the artistic intent as closely as possible”); and
modifying … the first set of control signals to generate a second set of control signals, wherein the second set of control signals corresponds to a retargeted motion output and suppresses vibration of the robotic system during the animation sequence (See at least Section 6, Page 7 “To retarget artist-specified input onto our physical robots, we seek to minimize differences between simulated and target states, putting a priority on the suppression of visible vibrations of large amplitude. To do so, we parameterize and optimize the time-varying motor angles, and solve for their optimal control” and cited portion above. At least an initial and final set of control signals or “points” are generated.
Examiner furthermore notes that the optimization process is clearly disclosed as iterative, as made clear by at least the caption of Figure 5 on Page 8 “Parameters are fitted with bisection, and we obtain the final parameters within 5 iterations”, Section 7 Page 12 “In our experiments, we set the relative residual tolerance to 10−4 and the maximum number of minimization iterations to 100” and the graph showing “iterations” adjacent); and
applying the second set of control signals to the robotic system to actuate the robotic system (See at least various figures comparing non-optimized, optimized, and fabrication (real) optimized trajectory comparisons. For example, Figures 7 – 9)
Hoshyari does not teach, but in combination with Hayashi teaches:
training a machine learning model based on the simulated vibration data (See at least [0045] “Furthermore, the computer 30 performs learning for reducing the vibration of the robot R, on the basis of the learning program (learning means) 33e, which is stored in the storage unit 33. For example, when the user inputs a target value for the vibration of the distal end section of the robot R and instructs start of the learning, by using the input device 34, the computer 30 moves the robot R in the simulation a plurality of times on the basis of the operation program 33b” and [0046] “Note that it is possible to use a well-known learning program and to achieve a reduction in the vibration of the distal end section of the robot R through various well-known learning techniques”);
…
[modifying], based on the machine learning model, [the first set of control signals to generate a second set of control signals] (See at least [0046] “Then, an improved operation program that is improved on the basis of the learning program 33e is stored in the storage unit 33”)
…
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to train and utilize a machine learning model such as that taught by Hayashi to perform the optimization taught by Hoshyari without the need to perform the optimization method each and every time to allow for faster or even on-line/real-time processing from a target input motion to a final retargeted motion, or alternatively and/or as part of such process, to utilize the optimization model of Hoshyari within the training of a machine learning model such as in Hayashi with a reasonable expectation of success. It is well understood and routine in the art of robotics to train and utilize machine learning models so that, once trained, faster or even on-line or real-time processing of data can be performed. It is similarly well understood and routine in the art of machine learning to utilize optimization functions so that a model is trained to provide desired and/or accurate results. Therefore, it would be obvious to one of ordinary skill in the art to extend the teachings of Hoshyari using machine learning models, particularly at the high level and generic level presently claimed, especially as the machine learning is vibration suppression specific as disclosed by Hayashi.
Regarding Claim 10, the combination of Hoshyari and Hayashi teaches:
The method of claim 8,
Hoshyari further teaches:
wherein training the machine learning model comprises modeling the robotic system as comprising a plurality of rigid components and a plurality of flexible components (See at least Page 2 “For our problem domain, techniques that couple different types of simulation models are highly relevant, since the physical characters we consider are multi-body systems composed of both rigid and flexible components”).
Regarding Claim 11, the combination of Hoshyari and Hayashi teaches:
The method of claim 8,
Hoshyari does not disclose, but Hayashi explicitly teaches:
further comprising receiving sensor data corresponding to movement of the robotic system based on the second set of control signals.
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to gather sensor data as taught by Hayashi in the system of Hoshyari with a reasonable expectation of success. The collection of sensor data is commonplace in the field of robotics, and the claim is not particular as to the nature of the sensor data (real, simulated, manner of collection, etc.). The information might be collected for any number of desirable reasons such as simple data logging or use in feedback loops.
Regarding Claim 12, the combination of Hoshyari and Hayashi teaches:
The method of claim 11,
Hoshyari does not disclose, but Hayashi explicitly teaches:
further comprising modifying, based on the machine learning model, the second set of control signals based on the sensor data to generate a third set of control signals corresponding to an updated retargeted motion output of the robotic system (See at least [0063] “The configuration of the vibration display device of the third embodiment is the same as that of the first embodiment. The operation program creating device of the third embodiment differs from that of the first embodiment in that the operation program creating device of the third embodiment has a vibration measurement sensor 52, as shown in FIG. 3, and the computer 30 acquires a vibration state of the distal end section of the robot R from the vibration measurement sensor 52” and [0045] which discloses an iterative nature of the first embodiment which is maintained per [0063] above, “For example, when the user inputs a target value for the vibration of the distal end section of the robot R and instructs start of the learning, by using the input device 34, the computer 30 moves the robot R in the simulation a plurality of times on the basis of the operation program 33b”).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to utilize sensor data and iterate learning as taught by Hayashi in the system of Hoshyari with a reasonable expectation of success. The claim is not particular as to the nature of the sensor data (real, simulated, manner of collection, etc.) or even the nature of the basis of modifying. Therefore, Hoshyari in combination with Hayashi clearly teaches the limitation, as Hayashi and Hoshyari both disclose iterative processing and Hayashi teaches sensor data collection for use in learning/training.
Regarding Claim 13, the combination of Hoshyari and Hayashi teaches:
The method of claim 11,
Hoshyari further teaches:
wherein the sensor data indicates changes in motion output of the robotic system over time (See at least [0035] “When the robot R is moved in the simulation, the computer 30 acquires a trajectory and a vibration state of the distal end section of the robot R. The vibration state to be acquired is, in one example, vibration data indicating vibration of the distal end section. The vibration data can be data of the acceleration of the distal end section, which changes over time, or data of the amplitude of the distal end section, which changes over time”).
Regarding Claim 14, the combination of Hoshyari and Hayashi teaches:
The method of claim 8,
Hoshyari further teaches:
wherein the training the machine learning model comprises:
providing a first set of vibration data generated by the differentiable simulator, wherein the differentiable simulator simulates the robotic system performing an animation based on an initial set of control signals comprising the first set of control signals (See at least Section 6, Page 7 “We initialize the control points by fitting the parameterized profiles to the input animation, collecting the spline parameters of all motors in a global parameter vector p”. Examiner notes that the phrase “vibration data” would appear to be any data relating in some manner to vibration, which would include any trajectory or control data. Examiner notes that Applicant does not appear to set forth a special definition for the noun phrase within the specification, or even appear to clearly and explicitly describe the noun phrase), wherein a first set of components of the robotic system are modeled as flexible components and a second set of components of the robotic system are modeled as rigid components (See at least Page 2 “For our problem domain, techniques that couple different types of simulation models are highly relevant, since the physical characters we consider are multi-body systems composed of both rigid and flexible components”);
providing a second set of vibration data generated by the differentiable simulator, wherein the second set of vibration data suppresses vibration of the robotic system as compared to the first set of vibration data (See at least Section 7, Page 12 “As we observe in the inset energy plot, our minimizations converge well and substantially decrease the vibrations from the input sequence, without sacrificing the quality of motion”. Each iteration is designed to further suppress vibration); and
generating a modified set of control signals based on differences between the first set of vibration data and the second set of vibration data (Examiner notes that this limitation appears to be particularly non-specific. The nature of “based on”, the “differences” themselves, and even the phrase “vibration data” are not claimed with any particularity. Examiner furthermore notes that the optimization process is clearly disclosed as iterative, as made clear by at least the caption of Figure 5 on Page 8 “Parameters are fitted with bisection, and we obtain the final parameters within 5 iterations”, Section 7 Page 12 “In our experiments, we set the relative residual tolerance to 10−4 and the maximum number of minimization iterations to 100” and the graph showing “iterations” adjacent. Therefore, at the present highly generalized level of claim construction, further iterations of both Hayashi and Hoshyari which are based on past iterations are inherently based on differences given that the term “differences”, “based on”, and “vibration data” are open to broad interpretation).
Regarding Claim 15, the combination of Hoshyari and Hayashi teaches:
The method of claim 14,
Hoshyari further teaches:
wherein the first set of vibration data and the second set of vibration data comprise time-varying motor values (See at least Section 6, Page 7 “We represent the time-varying angle θi(t, pi) of motor i (hereafter referred to as i’s motor profile) with a spline interpolation, parameterized with control points pi”) and simulation states (See at least Section 6, Page 7 “Hence, a first objective that comes to mind minimizes deviations of the generalized position vector U from its target, integrated over the interval [0,T ]”) of the first set of components and the second set of components based on the animation.
Regarding Claim 16, the combination of Hoshyari and Hayashi teaches:
The method of claim 8,
Hoshyari does not teach alone, but in combination with Hayashi as already discussed above teaches:
further comprising determining a time varying state of the robotic system and measuring an effectiveness of the machine learning model based on the time varying state (See at least Figures 7 – 10, 12, and 13 where the effectiveness of the methodology is evaluated by comparing the target positions and orientations and actual/absolute positions and orientations of the final optimized simulation/and or actually physically executed control scheme. While Hoshyari does not teach a machine learning model, Hoshyari does teach evaluating the methodology, which in the case of the combination of Hoshyari and Hayashi already discussed includes the machine learning model. Therefore, the nature of this combination is already presented at the end of Claim 8).
Regarding Claim 17, the combination of Hoshyari and Hayashi teaches:
The method of claim 16,
Hoshyari does not teach alone, but in combination with Hayashi as already discussed above teaches:
wherein measuring the effectiveness of the machine learning model comprises comparing differences in absolute positions or orientations of the robotic system to corresponding target positions or orientations (See at least Figures 7 – 10, 12, and 13 where the effectiveness of the methodology is evaluated by comparing the target positions and orientations and actual/absolute positions and orientations of the final optimized simulation/and or actually physically executed control scheme. While Hoshyari does not teach a machine learning model, Hoshyari does teach evaluating the methodology, which in the case of the combination of Hoshyari and Hayashi already discussed includes the machine learning model. Therefore, the nature of this combination is already presented at the end of Claim 8).
Regarding Claim 18, the combination of Hoshyari and Hayashi teaches:
The method of claim 8,
Hoshyari further teaches:
wherein the vibration that is suppressed by the second set of control signals consists of low-frequency, large-amplitude vibrations of the robotic system (See at least abstract of Page 1, “To this end, we develop an optimization-based, dynamics-aware motion retargeting system that adjusts an input motion such that visually salient low-frequency, large amplitude vibrations are suppressed”, Section 2, Page 2, “This setting introduces a challenge that has not yet been addressed in the computer graphics community, namely that of reasoning about visually salient low-frequency large amplitude structural vibrations”, and Section 6, Page 7, “putting a priority on the suppression of visible vibrations of large amplitude”, “In terms of frequencies and amplitudes, we can interpret our retargeting as compensating for low-frequency vibrations of large amplitude”, and “To avoid the “accumulation” of global and visible vibrations, and trade them for less visible vibrations”).
Regarding Claim 21, the combination of Hoshyari and Hayashi teaches:
The method of claim 8,
Hoshyari does not teach, but the combination of Hoshyari and Hayashi teaches:
wherein the machine learning model comprises comparisons of target states defined by a plurality of input animations defining motions for components of the robotic system and states determined by the differentiable simulator (See at least Page 7 of Hoshyari “To retarget artist-specified input onto our physical robots, we seek to minimize differences between simulated and target states, putting a priority on the suppression of visible vibrations of large amplitude” and “Our objective measures performance w.r.t. absolute coordinates. To provide a means to penalize relative differences, i.e., closeness to the artistic input, we formulate a regularizer that compares the current profile to the input profiles”, as well as [0005] found under “Background of the Invention” which discloses prior art and provides context to the disclosure of Hayashi, “learning is performed so as to bring the measured value close to a target value”, and [0045] “Furthermore, the computer 30 performs learning for reducing the vibration of the robot R, on the basis of the learning program (learning means) 33e, which is stored in the storage unit 33. For example, when the user inputs a target value for the vibration of the distal end section of the robot R and instructs start of the learning, by using the input device 34, the computer 30 moves the robot R in the simulation a plurality of times on the basis of the operation program 33b” and [0046] “At this time, the computer 30 searches for an optimal operation of the robot R while changing the operating speed, e.g., the operating speed of each of the servomotors 11a, 12a, 13a, 14a, 15a, and 16a, little by little. Then, an improved operation program that is improved on the basis of the learning program 33e is stored in the storage unit 33” of Hayashi.
It would be clear to one of ordinary skill in the art, particularly as the claim does not disclose with any particularity what the “comparisons” are or defines a “differentiable simulator” (though Hoshyari discloses a differentiable simulator), that the learning of the combination of Hoshyari and Hayashi involves comparisons between target/goal states and simulated states. Hoshyari clearly sets forth that optimization should occur by minimizing differences between target and simulated states. This should inherently require a comparison regardless of how it is implemented, otherwise it would be impossible to know how large or small the difference is, let alone minimize it. Additionally, Hayashi discloses the specific means of the regularizer which can be implemented within a learning model such as Hoshyari’s. Furthermore, Hayashi similarly discloses convergence on a target value even with their own unmodified learning, and additionally and separately even discloses incorporation of a human-in-the-loop evaluation for further comparison and determination of if additional optimization is required (See [0054] of Hayashi)
For these reasons, it is considered obvious that the combination of Hoshyari and Hayashi disclose these features).
Regarding Claim 22, the combination of Hoshyari and Hayashi teaches:
The method of claim 21,
Hoshyari further teaches:
wherein the differentiable simulator performs a dynamic simulation of the robotic system performing the retargeted motions with the components, and wherein the dynamic simulation predicts the vibration of the robotic system in performing the retargeted motion (See at least Page 5, Section 5 “Our dynamic simulation scheme, as described in the previous section, enables us to accurately simulate our robotic characters when actuating the motors according to the artist-specified motion profiles” and Figures 5 – 13).
Claim 20 is rejected under 35 U.S.C. 103 as being unpatentable over Hoshyari et al. in view of Hayashi et al. (US 20200338724 A1) and Mittal et al. (US 20210390288 A1).
Regarding Claim 20, the combination of Hoshyari and Hayashi teaches:
The method of claim 8,
The combination of Hoshyari and Hayashi does not explicitly teach, but Mittal teaches:
wherein the machine learning model comprises a neural network comprising layers of at least one of 1D convolutional networks or 1D recurrent neural networks (See at least [0079] “In certain embodiments, the method may also include processing the plurality of modalities via a plurality of 1D convolutional networks with batch normalization and a rectified linear activation function non-linearity, or a spatial temporal graph convolutional network”).
It would have been obvious to one of ordinary skill in the art prior to the effective filing date of the claimed invention to utilize of at least one of 1D convolutional networks in the machine learning model of the combination of Hoshyari and Hayashi as taught by Mittal with a reasonable expectation of success. The use of 1D convolutional networks or 1D recurrent neural networks are well known and routine the field of machine learning. Examiner notes that Applicant does not appear to disclose any particular basis for this architecture, and even discloses that it might be take on “a variety of forms” ([0154] of Applicant’s specification) which would appear to support Examiner’s position that it would be one of a variety of options available to one of ordinary skill in the art. Per prior art reference Zhu et al. (US 20210027379 A1), “one dimensional (1D) convolutional networks can be an effective tool to process time series” ([0037]), of which both Hoshyari and Hayashi are directed individually and in combination towards processing, which might serve as the motivation for one skilled in the art to select such layers for the architecture.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
White et al. (US 20200302241 A1) which discloses pre-training a machine learning model, including reasons for doing so. See for example [0024] “Training a machine learning model using simulated data has many advantages. A machine learning model pre-trained on simulated data requires less real-world training data and less time spent training on real-world data to achieve high-performance in new real-world environments compared to an untrained model.” White et al. also discloses further training using real sensor data after pre-training using simulated data [0025] “Using these tools, one can use simulation to generate large amounts of image data and corresponding labels that can be used to pre-train a ML model before it is trained further with real-world data or directly used for a real-world task” similar to [0142] of Applicant’s published Application.
Xu et al. (US 20200356849 A1) which discloses pre-training and no training during execution. See for example Figure 4.
Liebman et al. (US 20220034753 A1) which discloses pre-training a model for later use (with no further training). See for example [0041] “Further, in some implementations, for multiple different situations, conditions, or states of an engine based on a combination of sensor data inputs and operator control inputs—trained models as part of calibration data may be loaded or installed into a control module for an offline engine to calibrate the engine in accordance with one or more target objectives. In some implementations, in short, a simulation learning environment may include a simulator, where simulator output is a basis for developing trained models to be used in calibrating an offline engine. In some examples, a simulation learning environment may include an ensemble of multiple simulators” and [0131] “In some implementations, after one or more epochs of a genetic algorithm (510) and one or more rounds of optimization by an optimization trainer (560), the system (500) selects a particular model or a set of models as the final model (e.g., a model that is executable to perform one or more of the model-based operations of FIGS. 1-4B). … Subsequently, the final model can be output for use with respect to other data (e.g., real-time data)”.
Chau et al. (US 20220171373 A1) which discloses “The model is initially designed and trained offline using simulated data and then trained online using real tool data for predicting wafer routing path and scheduling” (Abstract).
Narayanan et al. (US 11816901 B2) which discloses pre-training a model without training during implementation.
Yoshida et al. (US 20220143823 A1) which discloses pre-training a model with simulated and data and further training using separate state data from running an operation using the trained model.
Handa et al. (US 20210122045 A1) which discloses motion retargeting using trained models, including pretraining models.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MATTHEW C GAMMON whose telephone number is (571)272-4919. The examiner can normally be reached M - F 10:00 - 6:00.
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, ADAM MOTT can be reached on (571) 270-5376. 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.
/MATTHEW C GAMMON/Examiner, Art Unit 3657
/ADAM R MOTT/Supervisory Patent Examiner, Art Unit 3657