01
Table for Two
A contact-physics testbed for turning language into coordinated two-arm manipulation. The scripted pipeline completed 29/30 unseen scenes; the learned policy solved early stages but not the full chain.一个将语言转化为双臂协同操作的接触物理试验台。脚本管线在未见场景中完成 29/30；学习策略能完成前期阶段，但未打通完整链路。
- Status状态
- Solo hackathon prototype独立黑客松原型
- Evidence证据
- 29/30 unseen scripted runs未见脚本运行 29/30
- Learned policy学习策略
- 0/10 complete task完整任务 0/10
- Runtime运行
- Local Intel CPU / iGPU本地 Intel CPU / iGPU
Read the case notes阅读案例摘要
Problem问题
Can low-cost home robots plan and execute two-handed everyday tasks on local hardware?低成本家用机器人能否在本地硬件上规划并执行双手日常任务？
My work我的工作
Language planner, validated task grammar, bimanual skills, contact tuning and evaluation scenes.语言规划、任务语法校验、双臂技能、接触参数调优与评估场景。
Proof验证
Physics-only manipulation, randomised scenes and OpenVINO inference without cloud or discrete GPU.纯物理操作、随机化场景，以及无需云端和独立显卡的 OpenVINO 推理。
