DiffMimic: Efficient Motion Mimicking with Differentiable Physics
Authors: Jiawei Ren, Cunjun Yu, Siwei Chen, Xiao Ma, Liang Pan, Ziwei Liu
ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on standard benchmarks show that Diff Mimic has a better sample efficiency and time efficiency than existing methods (e.g., Deep Mimic). Notably, Diff Mimic allows a physically simulated character to learn Backflip after 10 minutes of training and be able to cycle it after 3 hours of training, while the existing approach may require about a day of training to cycle Backflip. |
| Researcher Affiliation | Collaboration | Jiawei Ren 1 Cunjun Yu 2 Siwei Chen2 Xiao Ma3 Liang Pan1 Ziwei Liu1 1 S-Lab, Nanyang Technological University 2 School of Computing, National University of Singapore 3 SEA AI Lab |
| Pseudocode | Yes | Algorithm 1 Diff Mimic |
| Open Source Code | Yes | Our code is available at https://github.com/diffmimic/diffmimic. |
| Open Datasets | Yes | The motion clips are directly borrowed from AMP (Peng et al., 2021), which are originally collected from a combination of public mocap libraries, custom recorded mocap clips, and artist-authored keyframe animations. |
| Dataset Splits | No | The paper does not explicitly provide training, validation, or test dataset splits. It mentions evaluation metrics and rollout lengths but not the partitioning of a dataset. |
| Hardware Specification | Yes | For all the experiments, we run the algorithm with one single GPU (NVIDIA Tesla V100) and CPU (Intel Xeon E5-2680). |
| Software Dependencies | No | The paper mentions using Brax, Adam optimizer, and Swish activation, but it does not specify version numbers for these software components, which is required for reproducibility. |
| Experiment Setup | Yes | We use Adam optimizer (Kingma & Ba, 2014) in training, with a learning rate of 3e-4 that linearly decreases with training iterations. We apply gradient clipping and set the maximum gradient norm to 0.3. We set the batch size to the same as the number of parallel environments. Cyclic motions. We set the maximum iterations to 5000. The number of environments is set to 200. Acyclic motions. We set the maximum iterations to 1000. The number of environments is set to 300. |