Third-party notices
The BSD-3-Clause license at the repository root applies only to original ORCS material. Third-party material retains its upstream terms.
Adapted code
src/orcs/tasks/dodge/mdp/rewards.py::ball_clearance adapts MimicKit's
compute_dodge_reward from its SMP dodgeball environment.
- Source: xbpeng/MimicKit
- License: Apache-2.0; see
LICENSES/Apache-2.0.txt - Citation: Xue Bin Peng, MimicKit: A Reinforcement Learning Framework for Motion Imitation and Control, arXiv:2510.13794, 2025.
Research provenance
ORCS's virtual-assistance implementation and SMPL seed-state pipeline were written for this repository. They build on ideas described by the following works; no source code from them is bundled here:
- Mandi Zhao, Yifan Hou, Dieter Fox, Yashraj Narang, Ajay Mandlekar, and Shuran Song, DexMachina: Functional Retargeting for Bimanual Dexterous Manipulation, arXiv:2505.24853, 2025.
- Siheng Zhao, Yanjie Ze, Yue Wang, C. Karen Liu, Pieter Abbeel, Guanya Shi, and Rocky Duan, ResMimic: From General Motion Tracking to Humanoid Whole-body Loco-Manipulation via Residual Learning, arXiv:2510.05070, 2025.
External dependencies and data
The following projects are resolved separately and are not relicensed or bundled in the ORCS Python distribution:
| Dependency or data | Use | Upstream terms |
|---|---|---|
| SONIC via Mocke | Frozen motion prior and compatible contracts | Apache-2.0 code; NVIDIA Open Model License weights |
| GRAIL | Terrain-motion source data and staging tools | NVIDIA's upstream repository and data terms |
| OmniRetarget Dataset | Retargeted terrain and object motions | MIT as declared by its dataset card |
| SMPL-X | Manually supplied body-model files | SMPL-X model license |
| mjlab, RSL-RL, and the installed assets package | Simulation, learning, and model assets | Their respective distributions and licenses |
Users are responsible for obtaining restricted datasets and body models and for complying with their terms. A citation or acknowledgement does not replace a license.