| ReTime is a research extension of StarVLA, not an official StarVLA release. | |
| StarVLA: https://github.com/starVLA/starVLA | |
| Base commit: 3422b9f2387b6f682cf02802904a77b23ab13afd | |
| Copyright (c) StarVLA Team. | |
| The complete upstream LICENSE is retained, including its additional attribution | |
| and commit-preservation language. The original base commit and its preceding | |
| upstream commit remain separate in the supplied Git history. | |
| Modifications in this artifact include temporal sampling/proposals, ordinary | |
| importance correction, scoring utilities, model/data/trainer integration, | |
| portable configurations, and the ReTime command-line interface. | |
| Original ReTime additions in this artifact are distributed under the root | |
| LICENSE terms. Their authors' identities are withheld during anonymous review. | |
| Existing upstream author/copyright notices identify the respective third-party | |
| contributors. | |
| NVIDIA Isaac-GR00T-derived code retains its source notices and Apache-2.0 terms. | |
| See licenses/Isaac-GR00T-LICENSE. Modified source files carry a modification notice. | |
| https://github.com/NVIDIA/Isaac-GR00T | |
| OpenVLA/Prismatic-derived logging code retains its attribution. | |
| See licenses/OpenVLA-LICENSE (MIT). | |
| https://github.com/openvla/openvla | |
| The Florence wrapper attributes X-VLA-derived code in its source. | |
| See licenses/X-VLA-LICENSE (Apache-2.0). | |
| https://github.com/2toinf/X-VLA | |
| The minimal FrameSkip source snapshot retains third_party/frameskip/LICENSE. | |
| Rotation conversion code in starVLA/dataloader/gr00t_lerobot/transform/state_action.py | |
| derives from Diffusion Policy. Its MIT license is retained in | |
| licenses/Diffusion-Policy-LICENSE. | |
| The adapted msgpack-numpy utility in deployment/model_server/tools/msgpack_numpy.py | |
| retains the original BSD license in licenses/msgpack-numpy-LICENSE. | |
| Third-party components remain subject to their respective licenses; the root | |
| license does not replace these component-specific terms. | |
| No datasets, pretrained model weights, checkpoints, benchmark repositories, | |
| or robot-client contents are redistributed in the working source tree. | |