Mingyuan Ba

Robot learning, visuomotor imitation, tactile feedback, dexterous manipulation, and decentralized multi-robot collaboration.

Education

University of Sydney

Feb 2023 – Dec 2026 (expected)

Bachelor of Advanced Computing

Major in Computer Science; minor in Mathematics.

Research Experience

PAIR Lab, University of Sydney

Dec 2025 – Present

Honours Researcher · Supervisor: Weiming Zhi

Build Python- and ROS-based teleoperation and data-collection systems for robotic arms, dexterous hands, and visuotactile sensors. Develop multi-arm control and human-in-the-loop intervention, and deploy Diffusion Policy and ACT on real robots.

Zeno AI

Jul – Aug 2026

Embodied AI Intern · Hangzhou, China

Contributed to Zeno-1, a decentralized multi-robot embodied AI foundation model, and supported real-world collaboration demonstrations involving shared-object manipulation and deformable objects.

Paper

Touch Where It Matters: Learning Local Tactile Policies for Fine-Grained Gripper Control

Mingyuan Ba, Yanan Zhou, James Zhao, and Weiming Zhi^*

arXiv preprint

Paper

Summary

ContactPolicy uses tactile and force histories to regulate the gripper during contact-rich stages while a visual policy controls arm motion. An Intent policy manages control handover, enabling fine-grained adjustments and target retention during ongoing manipulation.

CoRE: Learning Collaboration-Role Experts for Decentralized Collaborative Manipulation with One Policy

Yanan Zhou, Zhaoyan Qian, Zihao Li, Mingyuan Ba, Ranpeng Qiu, and Weiming Zhi^*

arXiv preprint

Paper Project & demo

Summary

CoRE learns collaboration-role experts within one shared policy. Each robot acts from local visual observations and proprioception without inter-robot messages or role labels. Experiments study complementary behaviour across simulation and real-world manipulation tasks, including robustness to partner delays and slowdowns.

NestDex: Nested Policy Learning with Copilot Assisted Teleoperation for Dexterous Manipulation

James Zhao^†, Jinhe Tang^†, Mingyuan Ba, and Weiming Zhi^*

arXiv preprint

Paper Project & demo

Summary

A nested policy-learning framework that uses learned hand skills to assist dexterous demonstration collection, then trains an outer visuomotor policy for autonomous deployment.

Tri-Manual Visuomotor Imitation Learning of Robot Policies

James Zhao^†, Mingyuan Ba^†, and Weiming Zhi^*

arXiv preprint

Paper Project & demo

Summary

A tri-manual imitation learning system that lets one operator demonstrate behaviours for three robotic arms. The work introduces Dependency-Aware Tri-Arm Scheduling to retime demonstrations offline before training a synchronized policy.

^† Equal contribution. ^* Corresponding author.