BridgeVLA++ Training, Keyframe and Checkpoint Release Package
BridgeVLA++官方Hugging Face固定revision公开的混合发布包，共5,287个文件，包含预训练语料、关键帧缓存以及BridgeVLA/BridgeVLA++多套基准checkpoint。入口公开非gated且数据卡声明Apache-2.0，但RLBench、COLOSSEUM、GemBench、RMBench及上游模型资产仍按各自条款；本次核验确认文件树和下载入口，没有完成全量下载。
- 开放状态
- 项目制品开放·上游条款另适用
- 许可证
- Apache-2.0 declared for released package; third-party benchmarks, models and assets retain upstream terms
- 商业使用
- conditional on package scope and every upstream benchmark/model/asset license
- 人工评审结论
- verified_public_nongated_fixed_revision_mixed_training_package
- 机器人
- Franka Research 3、Dobot CR5A、simulated benchmark embodiments
- 任务
- BridgeVLA/BridgeVLA++ pretraining and five simulation/real-robot benchmark stacks; exact per-file task manifest must be reconstructed from repository scripts
- 模态
- RGB、3D scene features、language instructions、robot actions、keyframe caches、model checkpoints
- 格式
- mixed Hugging Face file tree plus repository download scripts; benchmark-native formats
- 采集/生成方式
- RLBench/PyRep and benchmark-specific engines
- 已知问题
- Mixed data-and-checkpoint package rather than a clean standalone corpus; full bytes and episode counts are not summarized; complete package not downloaded; third-party terms and environment closure remain substantial.