MuJoCo simulation, 190 simulated seconds shown in 18: lateral undulation, a range-triggered detour around a cylinder, and rejoining the route with no contact. Propulsion uses a simulated surrogate.
Slithering Robot Snake Prototype
A ten-link robot snake, about 0.95 m long, with nine yaw joints: a CAD assembly, robot description files, and a MuJoCo simulation where it slithers with lateral undulation, notices an obstacle with a range sensor, detours around it without touching it, and rejoins its route. Twelve tests check the geometry, joints, gait and clearance.
How it was made
Recorded: a creator-reviewed session, November 12, 2026.
Models
- gpt-5.6-sol · openai · execution · Codex
- gpt-5.6 · openai · execution · Codex
Agent
Codex
Pinned skills
- earthtojake/text-to-cad/skills/cad ↗Commit
fdbb4b4fb62d95ae298cfe9a46fdc7092bdaf423 - earthtojake/text-to-cad/skills/step-parts ↗Commit
fdbb4b4fb62d95ae298cfe9a46fdc7092bdaf423 - earthtojake/text-to-cad/skills/cad-viewer ↗Commit
fdbb4b4fb62d95ae298cfe9a46fdc7092bdaf423 - arpitg1304/robotics-agent-skills/skills/robotics-design-patterns ↗Commit
54f7b578f3dc269d29c0beb623b3f2611fd3a430 - arpitg1304/robotics-agent-skills/skills/robotics-testing ↗Commit
54f7b578f3dc269d29c0beb623b3f2611fd3a430
Tools and APIs
- MuJoCo ↗
Physics simulation of slithering and obstacle avoidance.
- NumPy
Gait and controller maths in the simulation scripts.
- pytest
Ran the 12 tests covering geometry, joints, gait and obstacle clearance.
- Web search
Looked up MuJoCo XML details (rangefinder sensors, contact and actuator settings, rendering).
- Image viewer
Inspected rendered CAD snapshots and simulation frames.
Ordered steps
Ask $possible for a robot snake
Invoke the $possible skill with the idea. It asks a few questions before preparing the build.
$possible I want to make a robot snake.Say what it should do
Say what it should do
I want it to slither convincingly and navigate around obstacles on its own.Describe the environment
Describe the environment
A smooth indoor floor. Continue from my previous answer.Say where you are starting from
Say where you are starting from
I am starting from scratch. I only have the idea.Choose the result you want
Answer its last question and accept the pack it proposes. The agent then builds the CAD, robot description, simulation and tests.
A convincing digital prototype I can inspect and simulate now, with a clear path toward fabricating it later. Yes, proceed with this pack.
About
A ten-link robot snake, about 0.95 m long, with nine yaw joints: a CAD assembly, robot description files, and a MuJoCo simulation where it slithers with lateral undulation, notices an obstacle with a range sensor, detours around it without touching it, and rejoins its route. Twelve tests check the geometry, joints, gait and clearance.
How it was made
An AI agent made this. Codex built it in one recorded session, starting from the $possible skill: it asked what to build, how it should behave, where, and what result was wanted, then prepared a build pack. The answers shown here came from a scripted test run, not typed live. The agent used the text-to-cad and robotics skills listed below, MuJoCo for simulation and pytest for checks. Possible captured the session with possible capture; tool outputs and file contents are left out.
What to know
This is a digital prototype. Propulsion in the simulation uses a surrogate, so real-world slithering, servo choice, friction, strength and manufacturability are unproven. The video and renders were regenerated from the session's own files after the originals were lost; link count, joints, tests and simulated distances match the session's report.
Exact prompt
Hidden characters, if present, appear as Unicode escapes$possible
I want to make a robot snake.
I want it to slither convincingly and navigate around obstacles on its own.
A smooth indoor floor.
Continue from my previous answer.
I am starting from scratch. I only have the idea.
A convincing digital prototype I can inspect and simulate now, with a clear path toward fabricating it later.
Yes, proceed with this pack.Technical details and provenance
- Provider
- openai
- Model
- gpt-5.6-sol
- Agent
- Codex
- Published
- Publication
- Made by Possible
- Source
- Original source ↗