Graph Networks as Learnable Physics Engines for Inference and Control
Authors: Alvaro Sanchez-Gonzalez, Nicolas Heess, Jost Tobias Springenberg, Josh Merel, Martin Riedmiller, Raia Hadsell, Peter Battaglia
ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Across seven complex, simulated physical systems, and one real robotic system (see Figure 1), our experimental results show that our GN-based forward models support accurate and generalizable predictions, inference models3 support system identification in which hidden properties are abduced from observations, and control algorithms yield competitive performance against strong baselines. |
| Researcher Affiliation | Industry | 1Deep Mind, London, United Kingdom. Correspondence to: Alvaro Sanchez-Gonzalez <alvarosg@google.com>, Peter Battaglia <peterbattaglia@google.com>. |
| Pseudocode | Yes | Algorithm 1 Graph network, GN |
| Open Source Code | No | The paper mentions "For full algorithm, implementation, and methodological details, as well as videos from all of our experiments, please see the Supplementary Material." and provides links to videos, but it does not explicitly state that the source code for their methodology is released or provide a direct link to a code repository. |
| Open Datasets | Yes | Our experiments involved seven actuated Mujoco simulation environments (Figure 1). Six were from the Deep Mind Control Suite (Tassa et al., 2018) Pendulum, Cartpole, Acrobot, Swimmer, Cheetah, Walker2d and one was a model of a JACO commercial robotic arm. |
| Dataset Splits | No | The paper mentions 'Random valid data' in Figure 5's caption and refers to 'data in the validation set' in the text discussing results, indicating that a validation set was used. However, it does not specify the exact split percentages or sample counts for the validation set used for reproducibility. |
| Hardware Specification | No | The paper does not provide specific details about the hardware (e.g., GPU model, CPU type, memory) used to run the experiments. It refers to 'simulated physical systems' and 'real robotic system' but not the computing infrastructure. |
| Software Dependencies | No | The paper mentions using 'Mujoco' and 'GRUs' but does not provide specific version numbers for any software dependencies, which would be necessary for reproducibility. |
| Experiment Setup | Yes | We swept over 20 unique hyperparameter combinations for the MLP architecture, with up to 9 hidden layers and 512 hidden nodes per layer. ... We used our GN-based forward model to implement MPC planning by maximizing a dynamic-state-dependent reward along a trajectory from a given initial state. ... For example, Figure 10 shows frames of simulated JACO trajectories matching a target pose and target palm location, respectively, under MPC with a 20-step planning horizon. |