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 .
Claims 1-3, and 6-11 are pending for examination. Claims 1, 7, and 8 are independent.
Response to Amendment
The office action is responsive to the amendments filed on 07/07/2026. As
directed by the amendments claims 1-3, and 6-11 are amended. Claims 4-5 are canceled.
Response to Arguments
Applicant's arguments filed 07/07/2026 have been fully considered but they are not fully persuasive.
Applicant arguments regarding 35 U.S.C. § 101:
Examiner response: Applicant’s arguments, with respect to 35 U.S.C. § 101 have been fully considered and are persuasive. The 35 U.S.C. § 101 rejection has been withdrawn.
Applicant arguments regarding 35 U.S.C. § 103:
Examiner response: Applicant’s arguments with respect to 35 U.S.C. § 103 have been considered but are moot because the new ground of rejection. An updated 35 U.S.C. § 103 rejection is provided below.
Claim Rejections - 35 USC § 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.
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) 1-11 is/are rejected under 35 U.S.C. 103 as being unpatentable over Gimelfarb et al. ("Contextual Policy Transfer in Reinforcement Learning Domains via Deep Mixtures-of-Experts", hereinafter "Gimelfarb") in view of Chua et al. ("Deep Reinforcement Learning in a Handful of Trials using Probabilistic Dynamics Models", hereinafter "Chua") and Ueda (US 20180354125 A1, hereinafter "Ueda").
Regarding Claim 1
Gimelfarb disclose: create an aggregate state transition model including a plurality of state transition models that predict a next state of the based on a measured state of the and a command for the ([Section 2 Transfer Learning and Section 3] In Section 3.2 describes continuous-state tasks with deterministic transitions
P
i
(
s
'
|
s
,
a
)
a
n
d
f
^
i
(
s
,
a
)
(i.e. state transition models that predict the next state), for task i, control policy with state s (i.e. measured state), and actions a (i.e. command or control action). Section 3 further defines aggregate state transition models based on the plurality of source transition models shown in section 3.2 and equation 6. [Section 4.2] describes cartPole control problem and discloses an object of control.), and the aggregate state transition model further configured to aggregate results of prediction by the plurality of state transition models; ([Section 2 Transfer Learning and Section 3] Section 3.1 further describes target dynamics and Equation 2
P
s
'
s
,
a
,
D
t
=
∑
i
=
0
n
P
i
^
s
'
s
,
a
a
t
,
i
(
s
)
which is produced by the dynamic models combined into an aggregate target dynamic prediction.)
execute, for each control period, respective processing of inputting the measured state of the generating a plurality of candidates for a command or a series of commands for the acquiring a plurality of states or series of states of the that are predicted from the state of the robot and the plurality of candidates for a command or a series of commands for the robot using the aggregate state transition model, deriving a reward corresponding to each of the plurality of states or series of states of the and, based on the derived rewards, generating and outputting a command that maximizes the reward; ([Section 2 Markov Decision Process, Section 3, Second page of Appendix] describes a iterative optimization algorithm (i.e. for each control period) for optimizing policy (i.e. deriving a reward) corresponding the plurality of states and outputting an action (i.e. command) that maximizes the reward.)
update the aggregate state transition model such that an error between a predicted next state of the corresponding to the output command, and a measured state of the corresponding to the next state, is reduced, ([Section 3 equation 1-6] describes a probability of observing a transition from state s to state s’ decreasing a function of the prediction error (i.e.
|
|
s
'
-
f
i
'
^
(
s
,
a
)
|
|
observed next state vs predicted next state) across task i (i.e. aggregate state transition model). This describes updating the mixture model and section 3.2 further teaches learning when stating “In principle, νi could be modeled as an additional output of the mixture model, νi(θ), and learned from data [4],”)
wherein the aggregate state transition model is a structure that consolidates outputs of the plurality of state transition models, in accordance with aggregating weights of the respective outputs, ([Section 3 equation 1-6] In equation 2 describes
P
s
'
s
,
a
,
D
t
=
∑
i
=
0
n
P
i
^
s
'
s
,
a
a
t
,
i
(
s
)
this summations shows
a
t
,
i
(
s
)
(i.e. weights of the respective outputs).)
wherein the ([Section 3 equation 1-6] describe the mixture weights
a
t
,
i
(
s
;
θ
)
are outputs of a mixture network and the parameter θ are updated by minimizing negative log likelihood.)
Gimelfarb does not explicitly disclose: A learning device for controlling a robot, comprising: control the robot based on the output command;
However, Chua discloses in the same field of endeavor: A learning device for controlling a robot, ([Section 7, Appendix A.6 – A.7, and Fig 2] describes machine learning in robot motion planning, evaluating DOF pusher/reacher tasks emulating robotic arm control. Examiner interprets the experiment involving a computer (i.e. Learning device).) comprising:
execute, for each control period, respective processing of inputting the measured state of the ([Section 6 and Algorithm 1] PETS begins every MPC time step from the current state st and performs a new prediction or optimization cycle at every t.), generating a plurality of candidates for a command or a series of commands for the robot ([Section 6 and Algorithm 1] Given current state st PETS considers action sequences at:t+T (i.e. candidate action sequence), acquiring a plurality of states or series of states of the robot that are predicted from the state of the robot and the plurality of candidates for a command or a series of commands for the robot using the aggregate state transition model, deriving a reward corresponding to each of the plurality of states or series of states of the robot, and, based on the derived rewards, generating and outputting a command that maximizes the reward; ([Section 5 first para, Section 6 and Algorithm 1] describes PETS evaluates expected trajectory reward over predicted trajectory and executing optimal action based on reward.)
control the robot based on the output command; ([Section 6, Algorithm 1, Section 7, Appendix A.6 – A.7, and Fig 2] at every MPC time step, PETS executes action at. The 7-DOF reacher/pusher experiments apply the resulting action to robotic arm control task.)
It would have been obvious to a person of ordinary skill in art before the effective filling date of the invention to implement the function of probabilistic ensembles with trajectory sampling disclosed by Chua into the method of Deep Mixtures-of-Expert disclosed by Gimelfarb to control a robot. The modification would have been obvious because one of the ordinary skills of the art would be motivated to utilize the feature of probabilistic ensembles with trajectory sampling disclosed by Chua as all the references are in the field of reinforcement learning. A person of ordinary skill of the art would have been motivated to perform the combination for being able to perform optimal actions for robotic arm control tasks.
Gimelfarb in view of Chua does not explicitly disclose: a processor configured to: respective processing of inputting the measured state of the robot received from one or more sensors,
However, Ueda discloses in the same field of endeavor: a processor configured to: ([Para 0028, 0093] discloses a processor)
respective processing of inputting the measured state of the robot received from one or more sensors, ([Para 0009, 0048, 0056, 0070, 0076 Fig 3 and Fig 6] describes reviing sensor data from a robot/machine and inputting the data.)
It would have been obvious to a person of ordinary skill in art before the effective filling date of the invention to implement the controller and machine learning device disclosed by Ueda into the method of Gimelfarb in view of Chua to control a robot. The modification would have been obvious because one of the ordinary skills of the art would be motivated to utilize the feature of the controller and machine learning device disclosed by Ueda as all the references are in the field of reinforcement learning. A person of ordinary skill of the art would have been motivated to perform processes robotic sensor data in reinforcement learning algorithm.
Regarding Claim 7
Gimelfarb in view of Chua and Ueda discloses: A learning method for controlling a robot, comprising ([Section 7, Appendix A.6 – A.7, and Fig 2] describes machine learning in robot motion planning, evaluating DOF pusher/reacher tasks emulating robotic arm control. Examiner interprets the experiment involving a computer.), by a computer: (Claim 7 is a method claim that corresponds to claim 1 and the rest of the limitations are rejected on the same ground)
Regarding Claim 8
Gimelfarb in view of Chua and Ueda discloses: A non-transitory computer-readable medium storing instructions learning program for controlling a robot ([Para 0028-0029, 0032], Ueda disclose non-volatile memory.), executable by a computer to perform (Claim 8 is computer-readable medium claim that corresponds to claim 1 and the rest of the limitations are rejected on the same ground)
Regarding Claim 2
Gimelfarb in view of Chua and Ueda discloses: The learning device of Claim 1, wherein, each control period, the processor generates one candidate for a command or series of commands for the robot,derives a reward that is based on the generated candidate, and updates, one or more times, the candidate for the command or series of commands such that the reward becomes larger, thereby generating a candidate for the command or series of commands. ([Section 2 Markov Decision Process, Section 3, Second page of Appendix], Gimelfarb describes an iterative optimization algorithm (i.e. for each control period) for optimizing policy (i.e. deriving a reward) corresponding the plurality of states and outputting an action (i.e. command) that maximizes the reward.)
Regarding Claim 3
Gimelfarb in view of Chua and Ueda discloses: he learning device of Claim 1, wherein, each control period, the processor generates a plurality of candidates for a command or a series of commands for the robot,and thereafter, acquires a state or a series of states of the robot that is predicted from each of the plurality of candidates. ([Section 3, equations 1-6, and Page 2 of Appendix] Gimelfarb describes acquiring states that are predicated based on the function.)
Regarding Claim 6
Gimelfarb in view of Chua and Ueda discloses: The learning device of Claim 1, wherein:the aggregate state transition model includes an error compensation model in parallel with the plurality of state transition models, andthe processor updates the error compensation model. ([Section 3, equations 1-6, and Page 2 of Appendix], Gimelfarb describes evaluating model error and updating parameters of the models.)
Regarding Claim 9
Gimelfarb in view of Chua and Ueda discloses: A control device, comprising:a memory configured to store an aggregate state transition model learned by the learning device of Claim 1; andwherein the processor is further configured to execute, for each control period, respective processing of inputting the measured state of the robot, generating a plurality of candidates for a command or a series of commands for the robot, acquiring a plurality of states or series of states of the robot that are predicted from the state of the robot and the plurality of candidates for a command or a series of commands for the robot using the aggregate state transition model, deriving a reward corresponding to each of the plurality of states or series of states of the robot, and, based on the derived rewards, generating and outputting a command that maximizes the reward. ([Section 2 Markov Decision Process, Section 3, Second page of Appendix] Gimelfarb describes an iterative optimization algorithm (i.e. for each control period) for optimizing policy (i.e. deriving a reward) corresponding the plurality of states and outputting an action (i.e. command) that maximizes the reward.)
Regarding Claim 10
Gimelfarb in view of Chua and Ueda discloses: A control method, comprising, by a computer:acquiring an aggregate state transition model from a memory that stores the aggregate state transition model learned by the learning device of Claim 1; andexecuting, for each control period, respective processing of inputting the measured state of the robot,generating a plurality of candidates for a command or a series of commands for the robot, acquiring a plurality of states or series of states of therobot that are predicted from the state of the robot and the plurality of candidates for a command or a series of commands for the robot using the aggregate state transition model, deriving a reward corresponding to each of the plurality of states or series of states of the robot,and, based on the derived rewards, generating and outputting a command that maximizes the reward. ([Section 2 Markov Decision Process, Section 3, Second page of Appendix] Gimelfarb describes an iterative optimization algorithm (i.e. for each control period) for optimizing policy (i.e. deriving a reward) corresponding the plurality of states and outputting an action (i.e. command) that maximizes the reward.)
Regarding Claim 11
Gimelfarb in view of Chua and Ueda discloses: A non-transitory recording medium storing a control program; that is executable by a computer to perform processing, the processing comprising:acquiring an aggregate state transition model from a memory that stores the aggregate state transition model learned by the learning device of Claim 1; andexecuting, for each control period, respective processing of inputting the measured state of the robot,generating a plurality of candidates for a command or a series of commands for the robot, acquiring a plurality of states or series of states of therobot that are predicted from the state of the robot and the plurality of candidates for a command or a series of commands for the robot using the aggregate state transition model, deriving a reward corresponding to each of the plurality of states or series of states of the robot,and, based on the derived rewards, generating and outputting a command that maximizes the reward. ([Section 2 Markov Decision Process, Section 3, Second page of Appendix] Gimelfarb describes an iterative optimization algorithm (i.e. for each control period) for optimizing policy (i.e. deriving a reward) corresponding the plurality of states and outputting an action (i.e. command) that maximizes the reward.)
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Hamaya et al. (US 12521876 B2) describes a robotic reinforcement learning system.
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 TEWODROS E MENGISTU whose telephone number is (571)270-7714. The examiner can normally be reached Mon-Fri 9:30-5:30.
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, ABDULLAH KAWSAR can be reached at (571)270-3169. 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.
/TEWODROS E MENGISTU/Primary Examiner, Art Unit 2127
Prosecution Insights