Overview

Mobile ALOHA (Fu, Zhao, Finn — Stanford, CoRL 2024) extends the stationary ALOHA teleoperation system with a wheeled mobile base for whole-body data collection. The operator physically backdrives two leader arms while the two follower arms mirror the motion, and moves with the base tethered at the waist, so arm and base trajectories are captured together. The system uses Interbotix arms (WidowX-250 6-DOF leaders, ViperX-300 6-DOF followers) on an AgileX Tracer-style base with consumer webcams and an onboard laptop GPU. It is a published open-hardware reference design, not a commercial SKU: the value is the whole-body teleop interface plus the co-training recipe that lets a handful of mobile demos, combined with stationary ALOHA data, learn tasks like cooking shrimp, calling an elevator, and rinsing a pan.

Highlights

• Approx. build cost USD 28,000-35,000 (paper reports ~$32k), positioned as a low-cost whole-body platform
• BOM: 2x Interbotix WidowX-250 6-DOF leader arms + 2x Interbotix ViperX-300 6-DOF follower arms + AgileX Tracer-style wheeled base + 3x Logitech webcams + onboard laptop GPU
• Whole-body teleoperation: one operator puppeteers both arms and drives the base at the same time
• Co-training with stationary ALOHA data improves task success by up to ~90%
• Demonstrated tasks: cooking, using an elevator, dishwashing/rinsing pans
• Open-source: Hardware Code (BOM + assembly) and ML Code (ACT imitation-learning training) released
• Origin: Stanford (Zipeng Fu, Tony Zhao, Chelsea Finn), CoRL 2024 (arXiv:2401.02117)

Specifications