DETAILED ACTION
Claims 32-51 are presented for examination.
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Drawings
The drawings received on 20 March 2023 are accepted.
Claim Objections
Claim 44 is objected to because of the following informalities:
Claim 44 second to last line recites “using the an edge embedding”. This appears to be typographic error for “using the [[an]] edge embedding”.
Appropriate correction is required.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale or otherwise available to the public before the effective filing date of the claimed invention.
Claims 32-39, 42-44, 47, and 49
Claims 32-39, 42-44, 47, and 49 are rejected under 35 U.S.C. 102(A)(1) as being anticipated by Li, Y., et al. “Learning Particle Dynamics for Manipulating Rigid Bodies, Deformable Objects, and Fluids” arXiv:1810.01566v2 (April 2019) (cited in IDS dated 27 November 2023) [herein “Li”].
Claim 32 recites “32. A method performed by one or more data processing apparatus for simulating a state of a physical environment.” Li page 14 section D disclose “The models are implemented in PyTorch.” PyTortch corresponds with a data processing apparatus.
Li abstract disclose:
Real-life control tasks involve matters of various substances—rigid or soft bodies, liquid, gas—each with distinct physical behaviors. This poses challenges to traditional rigid-body physics engines. Particle-based simulators have been developed to model the dynamics of these complex scenes; …. In this paper, we propose to learn a particle-based simulator for complex control tasks.
The particle based simulator for physical behaviors corresponds to simulating states of a physical environment.
Claim 32 further recites “the method comprising, for each of a plurality of time steps: obtaining data defining the state of the physical environment at the current time step.” Li page 6 section 4 disclose “We evaluate our method on four different environments containing different types of objects and interactions. We will first describe the environments and show simulation results.” The description of the environments including the objects and interactions correspond with data defining the environment for a time step.
Claim 32 further recites “generating a representation of the state of the physical environment at the current time step, the representation comprising data representing a graph comprising a plurality of nodes that are each associated with a respective current node embedding and a plurality of edges that are each associated with a respective current edge embedding.” Li page 6 section 4.2 second paragraph discloses “For FluidFall, we dynamically build the interaction graph by connecting each particle to its neighbors within a certain distance d.” Building the interaction graph correspond with generating a representation of the environment for a current time step.
Li page 4 section 3.2 second paragraph discloses “Dynamic graph building. The vertices of the graph are the union of particles for all objects …. The edges between these vertices are dynamically generated over R time to ensure efficiency and effectiveness.” The vertices corresponding with the particles is the graph comprising a plurality of nodes. The edges being dynamically generated corresponds with generating edges associated with a current edge embedding.
Claim 32 further recites “updating the graph at each of one or more update iterations, comprising, at each update iteration: processing data defining the graph using a graph neural network to update the current node embedding of each node in the graph and the current edge embedding of each edge in the graph.” Li page 4 section 3.2 second paragraph discloses “Dynamic graph building. The vertices of the graph are the union of particles for all objects …. The edges between these vertices are dynamically generated over R time to ensure efficiency and effectiveness.” Dynamically generating edges correspond with updating the graph at respective iterations.
Li page 4 third paragraph discloses “The encoders for objects are denoted as
f
O
e
n
c
and the encoder for relations as
f
R
e
n
c
.” The encoder for objects corresponds with a processing to update the current node embedding. The encoder for relations corresponds with a processing tup update the current edge embedding. See further Li page 4 equations (2) and (3) which use these encoders.
Claim 32 further recites “after the updating, processing the respective current node embedding for each node in the graph to generate a respective dynamics feature corresponding to each node in the graph.” Li page 7 figure 2 caption discloses “Qualitative results on forward simulation. We compare the ground truth (GT) and the rollouts from HRN (Mrowca et al., 2018) and our model (DPI-Net) in four environments (FluidFall, BoxBath, FluidShake, and RiceGrip). The simulations from our DPI-Net are significantly better.” Forward simulation of the DPI-Net model of the environments correspond with processing the respective network embedding to generate dynamics features. The results of the simulation correspond with dynamics features of the nodes of the graph.
Li page 15 first paragraph discloses “The output of the model is the 3 dimensional velocity, which is multiplied by Δt and added to the current position to do rollouts.” The 3D velocity is a specific example of a generated dynamic feature.
Claim 32 further recites “and determining the state of the physical environment at a next time step based on: (i) the dynamics features corresponding to the nodes in the graph, and (ii) the state of the physical environment at the current time step.” Li page 7 figure 2 shows an x-axis of time with a series of images depicting various states of the environment. Left-to-right correspond with subsequent time-steps and subsequent states of the environment.
Claim 33 further recites “33. The method of claim 32, wherein the data defining the state of the physical environment at current the time step comprises respective features of each of a plurality of particles in the physical environment at the current time step, and wherein each node in the graph representing the state of the physical environment at the current time step corresponds to a respective particle.” Li page 4 section 3.2 second paragraph discloses “Dynamic graph building. The vertices of the graph are the union of particles for all objects …. The edges between these vertices are dynamically generated over R time to ensure efficiency and effectiveness.” The vertices corresponding with the particles is the graph comprising a plurality of nodes where nodes represent particles of the physical environment. Li page 5 third bullet item disclose “We build edges dynamically, connecting a fluid particle to its neighboring particles.”
Claim 34 further recites “34. The method of claim 33, wherein the plurality of particles comprise particles included in a fluid, a rigid solid, or a deformable material.” Li title discloses “Learning Particle Dynamics for Manipulating Rigid Bodies, Deformable Objects, and Fluids.” Rigid bodies correspond with rigid solids. Deformable objects correspond with deformable material.
Claim 35 further recites “35. The method of claim 33, wherein for each of the plurality of particles, the features of the particle at the current time step comprise a state of the particle at the current time step, wherein the state of the particle at the current time step comprises a position of the particle at the current time step.” Li page 3 section 3.1 second paragraph disclose “the state of object i, containing its position
q
i
and velocity
q
˙
i
.” The object position corresponds with a position of the respective particle.
Claim 36 further recites “36. The method of claim 35, wherein for each of the plurality of particles, the state of the particle at the current time step further comprises a velocity of the particle at the current time step, an acceleration of the particle at the current time step, or both.” From the above list of alternatives Examiner is selecting “a velocity of the particle at the current time step.”
Li page 3 section 3.1 second paragraph disclose “the state of object i, containing its position
q
i
and velocity
q
˙
i
.” The object velocity corresponds with a velocity of the respective particle.
Claim 37 further recites “37. The method of claim 35, wherein for each of the plurality of particles, the features of the particle at the current time step further comprise a respective state of the particle at each of one or more previous time steps.” Li page 13 Appendix A “Control Algorithm” recites “Forward simulation using the current graph
G
^
t
+
1
←
Φ
G
t
.” Here, Gt is a previous time step compared to step t+1. Accordingly, the forward simulation determining respective features of the particle environment comprises at least one previous time step state.
Claim 38 further recites “38. The method of claim 35, wherein for each of the plurality of particles, the features of the particle at the current time step further comprise material properties of the particle.” Li page 3 section 3.1 second paragraph disclose “denotes its attributes (e.g., mass, radius).” At least mass is a material property of the respective particle object.
Claim 39 further recites “39. The method of claim 33, wherein generating the representation of the state of the physical environment at the current time step comprises generating a respective current node embedding for each node in the graph, comprising, for each node in the graph: processing an input comprising one or more of the features of the particle corresponding to the node using a node embedding sub-network of the graph neural network to generate the current node embedding for the node.” Li page 4 equation (4) discloses:
PNG
media_image1.png
200
400
media_image1.png
Greyscale
The object properties for node i at time step t+1 corresponds with processing input for each node. The encoder FO corresponds with a sub-network of the graph neural network used to generate a next/current node embedding.
Li page 3 section 3.1 second paragraph disclose “
o
i
=
x
i
,
a
i
O
, where
x
i
=
q
i
,
q
˙
i
is the state of object i, containing its position
q
i
and velocity
q
˙
i
.” The position and velocity correspond to respective features.
Claim 42 further recites “42. The method of claim 33, wherein each edge in the graph connects a respective pair of nodes in the graph, and wherein generating the representation of the state of the physical environment at the current time step comprises: identifying each pair of particles in the physical environment that have respective positions which are separated by less than a threshold distance; and for each identified pair of particles, determining that the corresponding pair of nodes in the graph are connected by an edge.” Li page 6 section 4.2 second paragraph discloses “For FluidFall, we dynamically build the interaction graph by connecting each particle to its neighbors within a certain distance d.” The certain distance d corresponds with a threshold distance.
Li page 5 third bullet item disclose “We build edges dynamically, connecting a fluid particle to its neighboring particles.”
Claim 43 further recites “43. The method of claim 33, wherein the current edge embedding for each edge in the graph is a predefined embedding.” Li page 4 equation (2) discloses “the encoder for relations fR to determine edges e of a current time step t based on previous influence objects h. The influence object of previous time steps are predefined embeddings with respect to a current time step t.
Claim 44 further recites “44. The method of claim 33, wherein generating the representation of the state of the physical environment at the current time step comprises generating a respective current edge embedding for each edge in the graph, comprising, for each edge in the graph:
processing an input comprising:
respective positions of the particles corresponding to the nodes connected by the edge,
a difference between the respective positions of the particles corresponding to the nodes connected by the edge,
a magnitude of the difference between the respective positions of the particles corresponding to the nodes connected by the edge,
or a combination thereof,
From the above list of alternatives Examiner is selecting “respective positions of the particles corresponding to the nodes connected by the edge.”
Li page 3 section 3.1 second paragraph disclose “
o
i
=
x
i
,
a
i
O
, where
x
i
=
q
i
,
q
˙
i
is the state of object i, containing its position
q
i
and velocity
q
˙
i
.” The position and velocity correspond to respective features.
Claim 44 further recites “using the an edge embedding sub-network of the graph neural network to generate the current edge embedding for the edge.” Li page 4 equation (2) discloses “the encoder for relations fR to determine edges e of a current time step t based on previous influence objects h. Using the encoder for relations to propagate influence from relations k (see Li page 4 ¶4) correspond with using previous edge relations to generate the current edge embedding (i.e.
e
k
,
t
l
).
Claim 47 further recites “47. The method of claim 35, wherein processing the respective current node embedding for each node in the graph to generate the respective dynamics feature corresponding to each node in the graph comprises, for each node: processing the current node embedding for the node using a decoder sub-network of the graph neural network to generate the respective dynamics feature for the node.” Li page 15 first paragraph discloses “The output of the model is the 3 dimensional velocity, which is multiplied by Δt and added to the current position to do rollouts.” The output 3D velocity corresponds with generating the respective dynamics feature for the nodes.
Li page 4 equation (4) discloses:
PNG
media_image1.png
200
400
media_image1.png
Greyscale
The object properties for node i at time step t+1 corresponds with processing input for each node. The encoder FO corresponds with a sub-network of the graph neural network used to generate a next/current node embedding including respective velocity of the objects.
Li page 3 section 3.1 second paragraph disclose “
o
i
=
x
i
,
a
i
O
, where
x
i
=
q
i
,
q
˙
i
is the state of object i, containing its position
q
i
and velocity
q
˙
i
.” The position and velocity correspond to respective features.
Claim 47 further recites “wherein the dynamics feature characterizes a rate of change in the position of the particle corresponding to the node.” Li page 3 section 3.1 second paragraph disclose “the state of object i, containing its position
q
i
and velocity
q
˙
i
.” The object velocity corresponds with a velocity of the respective particle. Li page 15 first paragraph discloses “The output of the model is the 3 dimensional velocity, which is multiplied by Δt and added to the current position to do rollouts.” 3D velocity corresponds with a rate of change in the position of the particle of the nodes.
Claim 49 further recites “49. The method of claim 47, wherein determining the state of the physical environment at the next time step based on: (i) the dynamics features corresponding to the nodes in the graph, and (ii) the state of the physical environment at the current time step, comprises: determining, for each particle, a respective position the particle at the next time step based on: (i) the position of the particle at the current time step, and (ii) the dynamics feature for the node corresponding to the particle.” Li page 15 first paragraph discloses “The output of the model is the 3 dimensional velocity, which is multiplied by Δt and added to the current position to do rollouts.” The 3D velocity corresponds with the dynamics feature. The current position corresponds with the position of the particle at the current time step. Multiplying the velocity by Δt and adding to the current position is a determining of a respective position based on the position of the particle at the current time step and the dynamics feature for the node correspond to the particle.
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.
Claims 40, 41, 45, and 46
Claims 40, 41, 45, and 46 are rejected under 35 U.S.C. 103 as being unpatentable over Li, Y., et al. “Learning Particle Dynamics for Manipulating Rigid Bodies, Deformable Objects, and Fluids” arXiv:1810.01566v2 (April 2019) (cited in IDS dated 27 November 2023) [herein “Li”] as applied to claim 39 above, and further in view of Sanchez-Gonzalez, A., et al. “Graph Networks as Learnable Physics Engines for Inference and Control” Proceedings 35th Int’l Conf. on Machine Learning (2018) (cited in IDS dated 27 November 2023) [herein “Sanchez-Gonzalez”].
Claim 40 further recites “40. The method of claim 39, wherein for each node in the graph, the input to the node embedding sub-network further comprises one or more global features of the physical environment.” Li does not explicitly disclose global features of the environment; however, in analogous art of graph network physics engines, Sanchez-Gonzalez page 2 section 3 second paragraph teaches:
We distinguish between static and dynamic properties in a physical scene, which we represent in separate graphs. A static graph Gs contains static information about the parameters of the system, including global parameters (such as the time step, viscosity, gravity, etc.), per body/node parameters (such as mass, inertia tensor, etc.), and per joint/edge parameters (such as joint type and properties, motor type and properties, etc.). A dynamic graph Gd contains information about the instantaneous state of the system.
Global parameters correspond with global features.
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Li and Sanchez-Gonzalez. One having ordinary skill in the art would have found motivation to use static graph of global parameters into the system of learning particle dynamics for the advantageous purpose to “generalize interaction networks” (see Sanchez-Gonzalez page 3 left column) and to maintain “accurate predictions from real and simulated data, and surprisingly strong and efficient generalization.” See Sanchez-Gonzalez abstract.
Claim 41 further recites “41. The method of claim 40, wherein the global features of the physical environment comprise forces being applied to the physical environment, a gravitational constant of the physical environment, a magnetic field of the physical environment, or a combination thereof.” From the above list of alternatives Examiner is selecting “a gravitational constant of the physical environment.”
Sanchez-Gonzalez page 2 section 3 second paragraph teaches:
We distinguish between static and dynamic properties in a physical scene, which we represent in separate graphs. A static graph Gs contains static information about the parameters of the system, including global parameters (such as the time step, viscosity, gravity, etc.), per body/node parameters (such as mass, inertia tensor, etc.), and per joint/edge parameters (such as joint type and properties, motor type and properties, etc.). A dynamic graph Gd contains information about the instantaneous state of the system.
A global parameter of gravity corresponds with a gravitational constant of the physical environment.
Claim 45 further recites “45. The method of claim 33, wherein at each update iteration, processing data defining the graph using the graph neural network to update the current node embedding of each node in the graph comprises, for each node in the graph: processing an input comprising: (i) the current node embedding for the node, and (ii) the respective current edge embedding for each edge that is connected to the node, using a node updating sub-network of the graph neural network to generate an updated node embedding for the node.” Li page 4 third paragraph discloses “The encoders for objects are denoted as
f
O
e
n
c
and the encoder for relations as
f
R
e
n
c
.” The encoder for objects corresponds with a processing to update the current node embedding. The encoder for relations corresponds with a processing tup update the current edge embedding. See further Li page 4 equations (2) and (3) which use these encoders.
Li page 14 appendix D further describes the propagation with the encoders in an MLP, but does not explicitly disclose the update of the current node embeddings has input of both the current node embeddings and the current edge embeddings.
However, in analogous art of graph network physics engines, Sanchez-Gonzalez page 3 left column teaches:
A core GN block (Figure 2b) contains three sub-functions—edge-wise,
f
e
, node-wise,
f
n
, and global,
f
g
—which can be implemented using standard neural networks. Here we use multi-layer perceptrons (MLP). … where
f
e
is first applied to update all edges,
f
n
is then applied to update all nodes, and
f
g
is finally applied to update the global feature.
Sanchez-Gonzalez page 3 Algorithm 1 further shows “compute node-wise features
f
n
g
,
n
i
,
e
^
i
” which shows inputs of the nodes n and edges e.
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Li and Sanchez-Gonzalez. One having ordinary skill in the art would have found motivation to use core GN blocks into the system of learning particle dynamics for the advantageous purpose to “generalize interaction networks” (see Sanchez-Gonzalez page 3 left column) and to maintain “accurate predictions from real and simulated data, and surprisingly strong and efficient generalization.” See Sanchez-Gonzalez abstract.
Claim 46 further recites “46. The method of claim 33, wherein at each update iteration, processing data defining the graph using the graph neural network to update the current edge embedding of each edge in the graph comprises, for each edge in the graph: processing an input comprising: (i) the current edge embedding for the edge, and (ii) the respective current node embedding for each node connected by the edge, using an edge updating sub-network of the graph neural network to generate an updated edge embedding for the edge.” Li page 4 third paragraph discloses “The encoders for objects are denoted as
f
O
e
n
c
and the encoder for relations as
f
R
e
n
c
.” The encoder for objects corresponds with a processing to update the current node embedding. The encoder for relations corresponds with a processing tup update the current edge embedding. See further Li page 4 equations (2) and (3) which use these encoders.
Li page 14 appendix D further describes the propagation with the encoders in an MLP, but does not explicitly disclose the update of the current edge embeddings has input of both the current node embeddings and the current edge embeddings.
However, in analogous art of graph network physics engines, Sanchez-Gonzalez page 3 left column teaches:
A core GN block (Figure 2b) contains three sub-functions—edge-wise,
f
e
, node-wise,
f
n
, and global,
f
g
—which can be implemented using standard neural networks. Here we use multi-layer perceptrons (MLP). … where
f
e
is first applied to update all edges,
f
n
is then applied to update all nodes, and
f
g
is finally applied to update the global feature.
Sanchez-Gonzalez page 3 Algorithm 1 further shows “compute output edges
f
e
g
,
n
s
j
,
n
R
j
,
e
j
” which shows inputs of the connected nodes n and edges e.
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Li and Sanchez-Gonzalez. One having ordinary skill in the art would have found motivation to use core GN blocks into the system of learning particle dynamics for the advantageous purpose to “generalize interaction networks” (see Sanchez-Gonzalez page 3 left column) and to maintain “accurate predictions from real and simulated data, and surprisingly strong and efficient generalization.” See Sanchez-Gonzalez abstract.
Dependent Claim 48
Claim 48 is rejected under 35 U.S.C. 103 as being unpatentable over Li, Y., et al. “Learning Particle Dynamics for Manipulating Rigid Bodies, Deformable Objects, and Fluids” arXiv:1810.01566v2 (April 2019) (cited in IDS dated 27 November 2023) [herein “Li”] as applied to claim 47 above, and further in view of Zhang, Y., et al. “FluidsNet: End-to-end learning for Lagrangian fluid simulation” Expert Systems with Applications, vol. 152, 113410 (April 2020) [herein “Zhang”].
Claim 48 further recites “48. The method of claim 47, wherein the dynamics feature for each node comprises an acceleration of the particle corresponding to the node.” Li page 9 section 4.2 last paragraph disclose “DPI-Nets, with state-specific motion predictors.” Li page 3 section 3.1 second paragraph teaches object positions and velocity. Li page 15 first paragraph discloses “The output of the model is the 3 dimensional velocity, which is multiplied by Δt and added to the current position to do rollouts.”
Li page 13 Algorithm 1 teaches “gradients” on a loss function of G to update the attributes.
But Li does not explicitly disclose an acceleration of the particle; however, in analogous art of graph neural network fluid simulation, Zhang page 3 algorithm 1 teaches a last loop “forall the particle i do” “
v
i
t
+
∆
t
=
v
i
t
+
∆
t
a
i
t
.” This teaches updating the velocity according to an acceleration
a
i
.
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Li and Zhang. One having ordinary skill in the art would have found motivation to use updating velocities into the system of learning particle dynamics for the advantageous purpose of updating the object velocities similar to the update of the object positions. Compare Zhang page 3 algorithm 1 and Li page 15 first paragraph.
Claims 50 and 51
Claims 50 and 51 are rejected under 35 U.S.C. 103 as being unpatentable over Li, Y., et al. “Learning Particle Dynamics for Manipulating Rigid Bodies, Deformable Objects, and Fluids” arXiv:1810.01566v2 (April 2019) (cited in IDS dated 27 November 2023) [herein “Li”] in view of US patent 11,455,374 B2 Belbute-Peres, et al. [herein “Belbute-Peres”].
Claim 50 recites “50. A system comprising: one or more computers; and one or more storage devices communicatively coupled to the one or more computers, wherein the one or more storage devices store instructions that, when executed by the one or more computers, cause the one or more computers to perform operations.” Li page 14 section D disclose “The models are implemented in PyTorch.” The use of PyTorch strongly suggests using a computer.
But Li does not explicitly disclose a computer; however, in analogous art of graph neural networks for fluid flow prediction, Belbute-Peres column 3 lines 13-15 teaches “data storage interface 180 may be a memory interface or a persistent storage interface, 30 e.g., a hard disk or an SSD interface.” Belbute-Peres column 3 lines 34-35 teach “The system 100 may further comprise a processor subsystem 160.” A system with a processor and memory corresponds with a computer and storage devices respectively.
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Li and Belbute-Peres. One having ordinary skill in the art would have found motivation to use a computer to execute the system of learning particle dynamics for the advantageous purpose of a hardware implementation which is art recognized suitable for use with graph neural networks. See generally Belbute-Peres column 3 line 7.
Claim 50 further recites “comprising: obtaining data defining the state of the physical environment at the current time step.” Li page 6 section 4 disclose “We evaluate our method on four different environments containing different types of objects and interactions. We will first describe the environments and show simulation results.” The description of the environments including the objects and interactions correspond with data defining the environment for a time step.
Claim 50 further recites “generating a representation of the state of the physical environment at the current time step, the representation comprising data representing a graph comprising a plurality of nodes that are each associated with a respective current node embedding and a plurality of edges that are each associated with a respective current edge embedding.” Li page 6 section 4.2 second paragraph discloses “For FluidFall, we dynamically build the interaction graph by connecting each particle to its neighbors within a certain distance d.” Building the interaction graph correspond with generating a representation of the environment for a current time step.
Li page 4 section 3.2 second paragraph discloses “Dynamic graph building. The vertices of the graph are the union of particles for all objects …. The edges between these vertices are dynamically generated over R time to ensure efficiency and effectiveness.” The vertices corresponding with the particles is the graph comprising a plurality of nodes. The edges being dynamically generated corresponds with generating edges associated with a current edge embedding.
Claim 50 further recites “updating the graph at each of one or more update iterations, comprising, at each update iteration: processing data defining the graph using a graph neural network to update the current node embedding of each node in the graph and the current edge embedding of each edge in the graph.” Li page 4 section 3.2 second paragraph discloses “Dynamic graph building. The vertices of the graph are the union of particles for all objects …. The edges between these vertices are dynamically generated over R time to ensure efficiency and effectiveness.” Dynamically generating edges correspond with updating the graph at respective iterations.
Li page 4 third paragraph discloses “The encoders for objects are denoted as
f
O
e
n
c
and the encoder for relations as
f
R
e
n
c
.” The encoder for objects corresponds with a processing to update the current node embedding. The encoder for relations corresponds with a processing tup update the current edge embedding. See further Li page 4 equations (2) and (3) which use these encoders.
Claim 50 further recites “after the updating, processing the respective current node embedding for each node in the graph to generate a respective dynamics feature corresponding to each node in the graph.” Li page 7 figure 2 caption discloses “Qualitative results on forward simulation. We compare the ground truth (GT) and the rollouts from HRN (Mrowca et al., 2018) and our model (DPI-Net) in four environments (FluidFall, BoxBath, FluidShake, and RiceGrip). The simulations from our DPI-Net are significantly better.” Forward simulation of the DPI-Net model of the environments correspond with processing the respective network embedding to generate dynamics features. The results of the simulation correspond with dynamics features of the nodes of the graph.
Li page 15 first paragraph discloses “The output of the model is the 3 dimensional velocity, which is multiplied by Δt and added to the current position to do rollouts.” The 3D velocity is a specific example of a generated dynamic feature.
Claim 50 further recites “and determining the state of the physical environment at a next time step based on: (i) the dynamics features corresponding to the nodes in the graph, and (ii) the state of the physical environment at the current time step.” Li page 7 figure 2 shows an x-axis of time with a series of images depicting various states of the environment. Left-to-right correspond with subsequent time-steps and subsequent states of the environment.
Claim 51 recites “51. One or more non-transitory computer storage media storing instructions that when executed by one or more computers cause the one or more computers to perform operations.” Li page 14 section D disclose “The models are implemented in PyTorch.” The use of PyTorch strongly suggests using a computer.
But Li does not explicitly disclose a computer; however, in analogous art of graph neural networks for fluid flow prediction, Belbute-Peres column 3 lines 13-15 teaches “data storage interface 180 may be a memory interface or a persistent storage interface, 30 e.g., a hard disk or an SSD interface.” Belbute-Peres column 3 lines 34-35 teach “The system 100 may further comprise a processor subsystem 160.” A data storage, memory, hard disk, and/or SSD correspond with a computer storage media storing instructions.
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Li and Belbute-Peres. One having ordinary skill in the art would have found motivation to use a computer to execute the system of learning particle dynamics for the advantageous purpose of a hardware implementation which is art recognized suitable for use with graph neural networks. See generally Belbute-Peres column 3 line 7.
Claim 51 further recites “comprising: obtaining data defining the state of the physical environment at the current time step.” Li page 6 section 4 disclose “We evaluate our method on four different environments containing different types of objects and interactions. We will first describe the environments and show simulation results.” The description of the environments including the objects and interactions correspond with data defining the environment for a time step.
Claim 51 further recites “generating a representation of the state of the physical environment at the current time step, the representation comprising data representing a graph comprising a plurality of nodes that are each associated with a respective current node embedding and a plurality of edges that are each associated with a respective current edge embedding.” Li page 6 section 4.2 second paragraph discloses “For FluidFall, we dynamically build the interaction graph by connecting each particle to its neighbors within a certain distance d.” Building the interaction graph correspond with generating a representation of the environment for a current time step.
Li page 4 section 3.2 second paragraph discloses “Dynamic graph building. The vertices of the graph are the union of particles for all objects …. The edges between these vertices are dynamically generated over R time to ensure efficiency and effectiveness.” The vertices corresponding with the particles is the graph comprising a plurality of nodes. The edges being dynamically generated corresponds with generating edges associated with a current edge embedding.
Claim 51 further recites “updating the graph at each of one or more update iterations, comprising, at each update iteration: processing data defining the graph using a graph neural network to update the current node embedding of each node in the graph and the current edge embedding of each edge in the graph.” Li page 4 section 3.2 second paragraph discloses “Dynamic graph building. The vertices of the graph are the union of particles for all objects …. The edges between these vertices are dynamically generated over R time to ensure efficiency and effectiveness.” Dynamically generating edges correspond with updating the graph at respective iterations.
Li page 4 third paragraph discloses “The encoders for objects are denoted as
f
O
e
n
c
and the encoder for relations as
f
R
e
n
c
.” The encoder for objects corresponds with a processing to update the current node embedding. The encoder for relations corresponds with a processing tup update the current edge embedding. See further Li page 4 equations (2) and (3) which use these encoders.
Claim 51 further recites “after the updating, processing the respective current node embedding for each node in the graph to generate a respective dynamics feature corresponding to each node in the graph.” Li page 7 figure 2 caption discloses “Qualitative results on forward simulation. We compare the ground truth (GT) and the rollouts from HRN (Mrowca et al., 2018) and our model (DPI-Net) in four environments (FluidFall, BoxBath, FluidShake, and RiceGrip). The simulations from our DPI-Net are significantly better.” Forward simulation of the DPI-Net model of the environments correspond with processing the respective network embedding to generate dynamics features. The results of the simulation correspond with dynamics features of the nodes of the graph.
Li page 15 first paragraph discloses “The output of the model is the 3 dimensional velocity, which is multiplied by Δt and added to the current position to do rollouts.” The 3D velocity is a specific example of a generated dynamic feature.
Claim 51 further recites “and determining the state of the physical environment at a next time step based on: (i) the dynamics features corresponding to the nodes in the graph, and (ii) the state of the physical environment at the current time step.” Li page 7 figure 2 shows an x-axis of time with a series of images depicting various states of the environment. Left-to-right correspond with subsequent time-steps and subsequent states of the environment.
Conclusion
Prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
US 11817184 B2 Park; Cheol Woo et al.
teaches
Graph neural network force field computational algorithms for molecular dynamics computer simulations
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Jay B Hann whose telephone number is (571)272-3330. The examiner can normally be reached M-F 10am-7pm EDT.
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, Renee Chavez can be reached at (571) 270-1104. 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.
/Jay Hann/Primary Examiner, Art Unit 2186 3 June 2026
Prosecution Insights