Forging Physical AI on AMD ROCm
在 AMD ROCm 上锻造物理智能
rocPAI-Forge is an open-source organization dedicated to building Physical AI solutions on AMD ROCm. We believe the next wave of intelligent systems will be grounded in the physical world — robots, simulation, perception, and action — and that a strong open ecosystem on ROCm is essential to make that future accessible to everyone.
Forge means to shape, refine, and build through practice. We don't just document APIs — we forge real solutions through hands-on engineering: training pipelines, simulators, deployment stacks, and reproducible workflows that push ROCm from capability to production-ready Physical AI.
Leverage AMD ROCm to advance Physical AI in the open: expand the ecosystem, share reference implementations, and turn research ideas into deployable systems — from simulation to the real world and back.
| Area | What We Explore |
|---|---|
| Sim2Real | Closing the gap between simulation and real robots — domain randomization, calibration, and deployment |
| Sim2Sim | Cross-simulator validation and transfer (e.g. MuJoCo ↔ custom backends) for robust policies |
| Real2Sim | Reconstructing scenes and dynamics from real data to improve simulation fidelity |
| 3D Assets & Scene Reconstruction | Meshes, environments, and digital twins for robotics and RL |
| Reinforcement Learning | Locomotion, manipulation, and task-specific RL on ROCm-accelerated stacks |
| World Models | Predictive models of environment dynamics for planning and control |
| VLA Models | Vision–Language–Action models for generalist robot policies |
| Real Robot Inference | Low-latency deployment on manipulators and mobile platforms with ROCm |
- Open by default — code, configs, and learnings shared with the community
- ROCm-first — optimize and validate on AMD hardware and software stack
- End-to-end — from data and sim to train, eval, and real-world inference
- Evidence over hype — reproducible benchmarks, clear contracts, and honest trade-offs
This organization is growing. Watch repos, open issues, and contribute PRs as projects land. For collaboration or questions, use Issues and Discussions in our repositories.
rocPAI-Forge 是一个专注于在 AMD ROCm 上构建 物理智能（Physical AI） 解决方案的开源组织。我们相信下一代智能系统将深深扎根于物理世界——机器人、仿真、感知与行动——而在 ROCm 上建设开放、可复现的生态，是让这一未来普惠开发者的关键。
Forge（锻造） 寓意通过实践去塑造、打磨与交付。我们不仅关注接口与文档，更通过工程实践锻造可落地的方案：训练管线、仿真栈、部署工具链，以及可复现的工作流，推动 ROCm 从「能用」走向「物理 AI 可投产」。
以 AMD ROCm 为底座，在开源社区推进物理智能：拓展生态、沉淀参考实现，把研究思路变成可部署系统——覆盖仿真到真机、真机回馈仿真的完整闭环。
| 方向 | 探索内容 |
|---|---|
| Sim2Real | 缩小仿真与真机差距——域随机化、标定与部署 |
| Sim2Sim | 跨仿真器验证与迁移（如 MuJoCo ↔ 自研后端），提升策略鲁棒性 |
| Real2Sim | 从真实数据重建场景与动力学，提升仿真保真度 |
| 3D 资产与场景重建 | 网格、环境与数字孪生，服务机器人与强化学习 |
| 强化学习 | 在 ROCm 加速栈上的运动、操作与任务型 RL |
| 世界模型 | 环境动力学预测模型，用于规划与控制 |
| VLA模型 | 视觉–语言–动作模型，面向通用机器人策略 |
| 真机推理 | 机械臂与移动平台上的低延迟 ROCm 部署 |
- 默认开源 — 代码、配置与经验向社区开放
- ROCm 优先 — 在 AMD 软硬件栈上优化与验证
- 端到端 — 从数据与仿真到训练、评测与真机推理
- 用结果说话 — 可复现基准、清晰契约、坦诚的技术取舍
组织与仓库持续建设中。欢迎 Star、提 Issue、参与 PR 与讨论。 合作与交流请通过各仓库的 Issues 与 Discussions 进行。
- Organization: github.com/rocPAI-Forge
- AMD ROCm: rocm.docs.amd.com
Forge the future of Physical AI on ROCm.
在 ROCm 上锻造物理智能的未来。