https://rocpai-forge.github.io/en/tags/rl/Recent content in RL on rocPAI-ForgeHugoenMon, 06 Jul 2026 00:00:00 +0000https://rocpai-forge.github.io/en/posts/openarm-rl-grasp/Mon, 06 Jul 2026 00:00:00 +0000https://rocpai-forge.github.io/en/posts/openarm-rl-grasp/<p><video src="https://rocpai-forge.github.io/media/openarm-rl-grasp/play_overview_web.mp4" poster="/media/openarm-rl-grasp/play_overview.jpg" autoplay loop muted playsinline style="width:100%;border-radius:.6rem;"></video></p> <p style="text-align:center;color:#888;font-size:.8rem;">总览回放 / Overview replay</p> <blockquote> <p>📖 This is the <strong>concise version</strong> (~3 min). For the full engineering details (design decisions, algorithm / reward, diagnostics, reproduce commands), read the <a href="https://github.com/rocPAI-Forge/tech-blog-pub/blob/main/PhysicalAI/openarm-rl-grasp/README-details.md"><strong>deep-dive →</strong></a></p></blockquote> <h2 id="unilab--joint-release">UniLab &amp; Joint Release</h2> <p><a href="https://github.com/unilabsim/UniLab">UniLab</a> is a <strong>heterogeneous robot-RL training infrastructure</strong>: <strong>CPU-parallel physics simulation</strong> (MuJoCo / Motrix) and <strong>GPU policy learning</strong> are coupled through a unified runtime and shared memory — instead of pinning physics, rollout collection, and learning on a single GPU-resident simulation path. Tasks, rewards, and backend selection are expressed as Hydra owner YAMLs; training goes through a unified <code>uv run train</code> / <code>uv run eval</code> CLI covering PPO, SAC, TD3, APPO, and more.</p>