hybrid · open-rua, dexjoco — Claude Code prompt for DexJoCo arm control
The hybrid Claude Code prompt controls an arm and dexterous hand assembly in DexJoCo. It supports pick_bucket with the published single-arm model recipe.
MIT ★ 488 updated 2026-11-15
.claude/commands/hybrid.mdWhat hybrid does
The prompt directs an agent to control an arm and hand in DexJoCo through the groups declared by actor_act. Each group couples the arm TCP with all sixteen hand joints in a single control step; this is not a scalar gripper. Instructions for actor_act must be concrete and bounded to the task.
The agent uses fresh scene observations between calls to decide whether to continue, revise the instruction, or finish. A completed chunk confirms that its control samples ran, not that the manipulation succeeded. The reviewer can allow a student prefix or stop and return control, but cannot edit TCP poses, fingers, or individual predictions.
For this binding, constraints must be empty; direct motion tools and EEF corrections are unavailable. The admitted task is pick_bucket with the published single-arm model recipe. Simulation pauses during inference and review, and unused predictions are discarded before the next inference.
What it helps with
- control the arm and every hand joint in one step
- choose whether to continue from fresh scene observations
- limit execution to the supported
pick_buckettask
Example prompts for the agent
- Run
pick_bucketin DexJoCo with the single-arm model recipe. - Continue the manipulation using fresh scene observations, then stop when it is complete.
- Give
actor_acta bounded instruction for the current scene and leaveconstraintsempty.
How to install hybrid
Manually: put the files in place
.claude/commands/hybrid.mdHow to use
Paste the text at the start of a chat or into project instructions
Questions about hybrid
How do I install hybrid in Claude Code?
Place the prompt file at .claude/commands/hybrid.md.
What does hybrid do?
It directs arm and sixteen-joint hand control in DexJoCo through actor_act, with scene observations used between calls.
Which task does hybrid support?
It admits only pick_bucket with the published single-arm model recipe. Similar names or a model’s general language abilities do not establish support for other tasks or robot layouts.
Source text
hybrid.md · MIT
DexJoCo provides one actor-controlled arm and dexterous hand assembly. Use the control groups declared by actor_act. Each group couples its arm TCP and all sixteen hand joints in one native control step. It is not a scalar gripper. Give actor_act a concrete, bounded local instruction consistent with the task. Use fresh scene observations between calls to decide whether to continue, refine the instruction or finish. A completed chunk means its control samples executed; it does not mean the requested manipulation succeeded. The reviewer can allow a student prefix or stop and return control. It cannot edit TCP poses, fingers or individual predictions. Direct motion tools and EEF corrections are unavailable. constraints must be empty for this binding. Samples represent simulated control time. Inference and review pause simulation; wall-clock latency is not native motion duration. Shorter prefixes reobserve more often. Unused predictions are discarded before the next inference. Only pick_bucket with the published single-arm model recipe is currently an admitted runtime. Do not infer support for other tasks or robot layouts from similar names or from a model's general language capabilities.
© Open-RUA · MIT · Open-RUA/uni-robot-agent
Similar prompts
All prompts for AI and ML →hybrid by other authors comes first.