hybrid · open-rua, isaaclab_arena — Claude Code prompt for humanoid control
The hybrid Claude Code prompt coordinates a humanoid as one actor. It combines upper-body targets with walking velocity while native control handles the legs and waist.
MIT ★ 488 updated 2026-11-15
.claude/commands/hybrid.mdWhat hybrid does
The prompt directs actor_act to send upper-body joint targets, planar walking velocity, and a pelvis-height reference as one coordinated action. Native whole-body control handles the legs and waist. The agent has no separate commands for those parts and no direct gripper tools.
Control calls must stay bounded, with a fresh head-camera observation checked before each new phase. The simulation advances only while acknowledged control samples execute. A preview shows requested controller references; it does not predict contact, balance, or motion. An acknowledged action alone does not prove a grasp or task completion. The task evaluator determines the official result.
A reviewer can allow a student prefix or stop the run, but cannot change whole-body geometry, joint targets, or locomotion references. constraints must be empty. Mentioning a skill in language instructions does not establish that the robot supports it.
What it helps with
- coordinate upper-body targets with walking velocity
- keep leg and waist control under the native controller
- distinguish an acknowledged action from task success
- check camera observations between control phases
Example prompts for the agent
- Control the humanoid according to the selected task and checkpoint.
- Inspect a fresh head-camera observation, then propose upper-body targets and walking velocity for the next phase.
- Stop if the next phase depends on a skill that has not been established as supported.
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?
Add hybrid.md to the project's .claude/commands/ directory.
What does the hybrid prompt do?
It coordinates humanoid actions through upper-body joint targets, walking velocity, and a pelvis-height reference. Native whole-body control maintains the legs and waist.
What are hybrid's control limits?
It provides no independent leg or waist commands and no direct gripper tools. Previews do not predict contact or balance, and the task evaluator determines the official outcome.
Source text
hybrid.md · MIT
The humanoid is one coordinated actor group. actor_act proposes upper-body joint targets together with planar walking velocity and a pelvis-height reference. Native whole-body control maintains the legs and waist. Neither direct gripper tools nor independent leg/waist commands are available. Use bounded actor calls and inspect fresh head-camera observations before deciding the next phase. Keep instructions consistent with the selected task and checkpoint. Language conditioning does not establish support for an unseen skill or task. A reviewer may allow a student prefix or stop. It cannot edit whole-body geometry, joint targets or locomotion references. constraints must be empty. The simulation advances only while executing acknowledged control samples. Thinking, inference and review do not consume simulated motion time. Previews show requested controller references, not predicted contact, balance or motion. An acknowledged action is not evidence of grasp or task success. The native task evaluator owns the official outcome independently of finish and review decisions.
© Open-RUA · MIT · Open-RUA/uni-robot-agent
Similar prompts
All prompts for AI and ML →hybrid by other authors comes first.