Robotics evidence guide · automated editorial preparation, not expert review

A reproducible grid-world pick-and-place experiment

A five-by-three grid represents carrying an object from the origin to a target. The direct controller moves horizontally then vertically. The search controller avoids blocked cells with a deterministic breadth-first search. Neither models grasp physics, vision, dynamics or real human assistance.

Both run the same six cases: clear path, detour, grasp failure, simulated assistance, insufficient action budget and offset target. The assistance event is a synthetic bookkeeping flag; it does not model an operator correcting the trajectory.

Computed result

direct-v1: 2/6 autonomous; 1 assisted; 3 failed. All 6 planned trials retained.

search-v2: 3/6 autonomous; 1 assisted; 2 failed. All 6 planned trials retained.

The difference is a result on deliberately selected fixtures—not an estimate of real-world robot reliability.

Execute locally

node --experimental-strip-types scripts/robotics-simulation.ts summary
node --experimental-strip-types scripts/robotics-simulation.ts package
node --experimental-strip-types scripts/robotics-simulation.ts verify <local-json-file>

The command does not contact a service or retain data. Save the package output locally to use the verify command. A refusal exits with a nonzero status.

What remains untested

  • No physical robot tested
  • No physics or perception validation
  • No population generalization
  • No safety certification
  • No ROS or LeRobot adapter implemented
Inspect the package specification