Get Started
Research questionHow should predictive models for robot feedback control be evaluated when measurement updates alter closed-loop behavior?A model that predicts accurately without new measurements may not identify the best controller once a robot repeatedly senses, updates its state, and acts again. This makes it difficult to use open-loop prediction scores as proxies for feedback performance.
Evaluation & Benchmarks
Machine Learning
Robotics
Latest papersRecent research connected to this question, newest first.Do Better Imagined Rollouts Mean Better Robot Control? A Controlled Study of World-Model Evaluation Under FeedbackEvidence comes from a differential-drive path-tracking task with biased odometry and intermittent landmark sensing. Six state estimators were tested across 24 sensing conditions using trajectory replay, 20-step measurement-free rollouts, and closed-loop tracking, with additional variation in rollout horizon, measurement-update interval, and recurrent training for sensing outages. The findings are limited to these controlled conditions and do not establish broader performance across robot platforms or tasks.research paper · Sep 2, 2026
Related questions
How can model predictive control remain feasible and stable for nonlinear systems with intermittent state measurements?How can text-based world models be evaluated for behavioral fidelity in long-horizon agent planning?How can autonomous-driving planners be stress-tested in realistic closed-loop scenarios that expose failures missed by nominal benchmarks?How can real-time MPC enforce control constraints and recover from infeasible initializations in strongly underactuated legged robots?
Home
Topics
Search
Library