Zewen (Anthony) Chen

Zewen (Anthony) Chen

Master of Mechanical Engineering at Columbia University, expert in Robotics and Control

MiniBEE 👌🐝: Towards a Miniature Bimanual End-Effector for Compact Coordinated Robot Dexterity

Minibee Design

Manipulation Tasks

Conducted in the ROAM Lab, MiniBEE is a bi-manual end-effector composed of two small robotic arms—each equipped with a gripper—mounted onto a larger industrial arm to enable mobile dexterous manipulation within an extended workspace. MiniBEE features:

  • A novel 8-DOF robot arm design with two end-effector, which enables dexterous bimanual manipulation and can be mounted on a larger industrial robot arm
  • Workspace analysis tools for evaluating dexterous manipulation under different kinematic designs
  • Kinematic control algorithms, including null-space control and sample-based singularity avoidance
  • A multifunctional interactive graphical interface supporting individual joint position control, Cartesian velocity control, motor torque enable/disable, trigger-to-gripper mapping, camera preview, etc.
  • A customized wearable back brace for mounting MiniBEE and collecting data in diverse environments
  • A demonstration collection and replay pipeline using puppeteering manipulation
  • Automatic policy learning for multiple dexterous manipulation tasks using Diffusion Policy

This project was exhibited at MIT NEMS 2025 with a poster and live demo, and our manuscript MiniBEE: A New Form Factor for Compact Bimanual Dexterity has been submitted to ICRA 2026, in which I am a co-first author.

My contributions include:

  1. Hardware improvement and assembly
  2. Development of kinematic control algorithms
  3. Design of the interactive graphical user interface
  4. Construction of the data collection pipeline
  5. Training and fine-tuning the Diffusion Policy for multiple dexterous manipulation tasks
  6. Collaboration in preparing the manuscript and supplemental videos

After ICRA 2026 submission, starting form October 2025, we are now focus on improving MiniBEE from following aspects:

  1. Residual RL for compensating the steady-state error in the motor command, reducing the gap between puppeteering and replay
  2. Speeding up the BC policy using RL
  3. Analysis of the “Continuous” Dexterous Workspace of different kinematic designs
  4. Environmental-related policy for closed-loop pick up and placement