Robotics planningAdvanced

Learned Feasibility Prediction for Robot Plans

A feasibility predictor estimates which symbolic actions are likely to survive expensive motion checks.

TAMPLearningFeasibilityRobotics

Site connection

The TAMP project trains feasibility guidance from successful and failed traces while keeping motion checking as the verifier.

Visual model

Guidance before verification

The predictor ranks or prunes plans before IK, collision, and trajectory checks spend real compute.

Interactive

Learning can rank plans, but geometry still verifies them

1Symbolic skeletonplanner space
2Arm assignmentplanner space
3IK checkmotion verifier
4Collision checkmotion verifier
5Trajectorymotion verifier
6Trace logtraining data

The Supervision Signal

Every failed IK call and collision check is useful data. A trace tells the learner what kind of partial plan was plausible symbolically but bad geometrically.

What It Should Not Do

The predictor should not declare a plan safe. It should only help decide what to try first or what to avoid trying too often.

Common Pitfalls

  • Training only on successful traces.
  • Letting the predictor replace collision checking.
  • Optimizing median planning time while ignoring rare catastrophic failures.

Quick check

Quiz

What is the final verifier in this architecture?
  1. The learned predictor
  2. IK, collision, trajectory, and execution checks
  3. A CSS animation
  4. The task name

Learned guidance helps search, but geometric and execution checks verify feasibility.

Sources and Further Reading

Related Explainers