Sankalp Yamsani

This work adds one feature-alignment term to ordinary VLA training: a frozen world model is run over the training frames once and cached, and the student learns to agree with that cache, indicating a broad representational prior rather than a fragile alignment between two particular networks.

Achieving 89.12% accuracy in cluttered kitchen environments, this modular pipeline enables robots to adapt seamlessly to diverse household tasks without retraining.

Collect high-quality robot manipulation data anywhere with TRIP-Bag, a teleoperation system that fits in a suitcase and sets up in under 5 minutes.