Internship - Physical AI for Dexterous, Contact-Rich Humanoid Manipulation

Internship - Physical AI for Dexterous, Contact-Rich Humanoid Manipulation Mitsubishi Electric Research Laboratories (MERL) is seeking a highly motivated Ph.D. student to develop hybrid learning and control methods for reliable humanoid manipulation in factory automation. The research will combine Vision-Language-Action (VLA) models and World-Action Models (WAMs) with classical motion planning, feedback control, and perception. A central question is how robots should coordinate these components, recognize their limitations, and recover when execution deviates from the intended task. The intern will develop and evaluate novel algorithms on physical robotic platforms, targeting industrial tasks such as container stacking and transport, kitting, tool manipulation, and assembly/disassembly. The project will combine algorithm development with simulation and real-world experiments, comparing approaches through task success, recovery performance, execution time, and constraint violations. The intern will work closely with MERL researchers on a focused research contribution, with the goal of submitting findings to leading robotics or machine-learning conferences such as CoRL, RSS, ICRA, IROS, NeurIPS, or ICML. Required Specific ExperienceCurrent enrollment in a Ph.D. program in robotics, computer science, mechanical engineering, electrical engineering, or a related field. Hands-on experience adapting or evaluating VLA models, such as GR00T, p0.5, or related models, including deployment on physical robotsStrong foundations in robot manipulation, motion planning, and feedback control. Proficiency in Python, ROS2, and a deep-learning framework such as PyTorch. Experience conducting robotic experiments using simulation environments such as Isaac Sim or MuJoCo. Demonstrated research ability through publications, substantial research projects, or research software. Preferred ExperienceHumanoid (Unitree G1, Agibot G2) or bimanual manipulation, including integration with whole-body control. World-action models or learned world models for prediction, planning, or control. Learning from demonstrations, teleoperation, or autonomous data collection. Uncertainty estimation, failure detection, policy intervention, or sim-to-real transfer. Contributions to open-source robotics or robot-learning repositories. The pay range for this internship position will be 6-8K per month.