Physical Intelligence for Robotics

This research project focuses on the development of advanced algorithms and AI models for physical intelligence. By integrating multimodal perception, natural language understanding, and intelligent manipulations on the Mobile ALOHA platform, we aim to achieve autonomous and intelligent execution of complex manipulation tasks in unstructured real-world environments.

Projects

Collecting and Scheduling Demonstrations for Multi-Step Precision Manipulation with VLAs, IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). 2026.

A complete pipeline for applying vision-language-action (VLA) models to multi-step precision manipulation, and a systematic study of how those demonstrations should be scheduled during fine-tuning.

Leader-Free Mobile ALOHA: Gamepad-IK Teleoperation and VLA Fine-Tuning for Laboratory Manipulation

A leader-arm-free teleoperation system that uses a standard gamepad with inverse kinematics (IK) to control the full Mobile ALOHA platform—both bimanual arms and the mobile base.

Real2Sim2FEA: From Robot Task Data to Structural Load Spectra

A general method for turning a robot’s recorded task data into load cases and cycle spectra for its own structure — then that method narrowed onto one bimanual ALOHA task, 557 episodes of it, and carried through to a solved upper-arm stress field. (ALOHA Mobile → MuJoCo → Ansys)

Hardware and Compute

Title: “Fine-tuning Foundation Models for Robotic Manipulation: Advancing VLA Models through Domain-Specific Adaptation”; Source of Support: NSF ACCESS; Award Number: #CIS251342 (750,000 GPU Computing Credits)

Title: “Immersive Teleoperation for Mobile ALOHA: Advancing Vision-Language-Action Models through Spatial Computing”; Source of Support: Spatial Computing Hub at Purdue University; Support: Apple Vision Pro device