Leader-Free Mobile ALOHA

Abstract

We present 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. Unlike the standard leader-follower interface, which physically tethers the operator to the robot, our gamepad-IK approach decouples the operator, enabling them to walk close to the robot for visual precision and move freely around the workspace to find optimal viewing angles. We demonstrate this system on three laboratory manipulation tasks including precision pipette tip attachment, tube pick-and-place, and bimanual pipette tip release, showcasing the system's capability for both single-arm and coordinated bimanual control. Using 55 minutes of gamepad-collected demonstrations, we fine-tune the π0.5 VLA and show that the resulting policy can correctly select the appropriate arm based on object position and successfully execute tube pickup, with robust visual tracking of object positions during execution. Code, videos, and dataset are available at: https://shengfeng-yang.github.io/aloha-gamepad/.

Teleoperation Demonstrations

Leader-free Teleoperation

Tube Pick-and-Place

Pipette Tip Pickup

Bimanual Pipette Tip Release

Autonomous Policy (π0.5) Demonstrations

Arm Selection (Right Arm)

Distractor Robustness

Visual Tracking (Move Rack & Tube)

Visual Tracking (Move Rack Twice)

Key Idea: Leader-Free Operator Mobility

Leader-follower teleoperation tethers the operator to the robot. For precision tasks like laboratory manipulation, this limits the operator's ability to see fine contact details. By replacing leader arms with a wireless gamepad + IK, the operator can walk freely—getting close for precision, orbiting for better angles, and stepping back for a global view.

❌ Standard (Leader-Follower)

Operator fixed behind robot
~50 cm from end-effector
No viewpoint adjustment

✅ Leader-Free (Ours)

Operator free to move anywhere
5–15 cm from contact point
Walk-around viewpoint optimization

Method

Our system maps gamepad inputs to Cartesian end-effector velocities, solved via inverse kinematics at 50 Hz for each arm of the Mobile ALOHA platform. Demonstrations are recorded in LeRobot v3.0 format and used to fine-tune π0.5 for autonomous execution.

System pipeline: gamepad teleoperation, data collection, pi0.5 fine-tuning, autonomous policy — System pipeline. Phase 1: the operator uses a PS4 gamepad to control the Mobile ALOHA via inverse kinematics and records demonstrations. Phase 2: the dataset is used to fine-tune π0.5 for autonomous execution.

Gamepad Control Mapping

The controller uses a modifier-based design that balances workload across both hands. Position control (left stick) and height/roll (right stick) are distributed by default, while L2 and R2 modifiers allow either hand to temporarily take on base or orientation control.

Safety Features

The system includes workspace limits (distance-dependent Cartesian bounds), IK fallback with automatic recovery after consecutive failures, per-joint limit validation with safety buffers, and acceleration-limited base control behind an L2 deadman switch.

Results

We evaluate on a tube pick-and-place task using the Mobile ALOHA platform. The tube and rack are placed at varying positions near either the left or right arm, requiring the policy to select the correct arm and execute the grasp. The VLA model is fine-tuned on 55 minutes of gamepad-collected demonstrations.

Capability	Observation
Arm selection	Reliably selects the closest arm based on visual observation
Tube pickup	Successfully grasps tubes from rack in majority of trials
Visual tracking	Dynamically adjusts trajectory when tube is moved during execution
Distractor robustness	Correctly identifies target tube despite novel background objects not seen during training

BibTeX

@misc{yang2026aloha-gamepad,
  title={Leader-Free Mobile ALOHA:
         Gamepad-IK Teleoperation and VLA Fine-Tuning
         for Laboratory Manipulation},
  author={Yang, Shengfeng},
  year={2026},
  url={https://shengfeng-yang.github.io/aloha-gamepad/}
}