Mobile ALOHA | sudoremove

Key Significance

Democratization of Mobile Bimanual Manipulation: Implemented ~$32,000 platform vs existing $200,000+ mobile manipulators - greatly improved research accessibility
Whole-Body Teleoperation: Enables complex household tasks that require moving while using both arms, beyond tabletop manipulation
Co-Training Paradigm: Achieves 80-90% success rate with only 50 demonstrations by jointly training with existing static ALOHA datasets
Open-Source Ecosystem: Complete disclosure of hardware design, software, 3D printing files, assembly tutorials
Practical Household Robot Research: Demonstrated potential for general household robots through real-life task demonstrations like cooking, cleaning, elevator calling

Overview

Mobile ALOHA is a low-cost whole-body teleoperation system developed by Stanford IRIS Lab. By mounting the existing tabletop-only ALOHA system on a mobile base, it enables learning complex tasks that simultaneously perform locomotion and bimanual manipulation.

Item	Spec
Development	Stanford IRIS Lab
Authors	Zipeng Fu, Tony Z. Zhao, Chelsea Finn
Publication	arXiv: January 2024 / CoRL 2024 presentation
Base	AgileX Tracer AGV
Arms	ViperX 300 6-DoF x 2 (follower) + WidowX 250 x 2 (leader)
Gripper	Custom parallel gripper (3D printed)
Cameras	Wrist x 2 + front 1 (Logitech C922x)
Compute	Laptop (RTX 3070 Ti, i7-12800H)
Total Cost	~$32,000
Paper	arXiv:2401.02117
Project	mobile-aloha.github.io

Hardware Configuration

Mobile Base: AgileX Tracer AGV

Source: AgileX TRACER Documentation

Item	Spec
Drive	2-wheel differential drive + 4 freewheel casters
Motors	150W brushless servo x 2
Max Speed	1.6 m/s (human walking speed)
Payload	100 kg
Size (L x W x H)	660 x 516 x 163.5 mm
Ground Clearance	30 mm
Obstacle Crossing	10 mm height, 8 degree slope
Operating Time	Up to 4 hours (100kg load)
Price	~$7,000

Robot Arms: ViperX 300 6-DoF

Source: Trossen Robotics ViperX 300

Item	Spec
Configuration	2 followers (for autonomous execution)
Arm DoF	6-DoF (arm body)
Gripper	1-DoF (open/close)
Arm+Gripper Total DoF	7-DoF per arm
Horizontal Reach	75 cm (base center to gripper)
Total Span	150 cm
Working Payload	750 g
Servos	DYNAMIXEL XM540-W270-R, XM430-W350-R
Resolution	4096 positions
Material	20mm x 40mm extruded aluminum
Price	~$6,130 x 2

Teleoperation Leader Arms: WidowX 250 6-DoF

Item	Spec
Configuration	2 leaders (for data collection)
Features	3D printed ergonomic handles
Use	Used only for demonstration data collection
Price	~$3,550 x 2

Sensors and Compute

Item	Spec
Cameras	Logitech C922x x 3
Resolution	640 x 480
Control Frequency	50 Hz (camera streaming and policy execution)
Placement	2 wrist + 1 front
Compute	Consumer-grade laptop
GPU	NVIDIA RTX 3070 Ti (8GB VRAM)
CPU	Intel i7-12800H
Communication	USB serial (arms) + CAN bus (base)

Power System

Item	Spec
Battery	1.26 kWh
Weight	14 kg
Position	Bottom of base (doubles as counterweight)
Features	Untethered wireless operation

Base Weight

Item	Spec
AgileX Tracer Weight	30 kg
Battery Weight	14 kg

Cost Breakdown (~$32,000)

Source: Mobile ALOHA Project Page, Trossen Robotics

Component	Price (USD)	Notes
AgileX Tracer AGV	~$7,000	Mobile base
ViperX 300 6-DoF x 2	~$12,260	Follower arms
WidowX 250 6-DoF x 2	~$7,100	Leader arms (for teleop)
Battery (1.26kWh)	~$2,000	Estimate
Cameras (C922x x 3)	~$300	RGB webcams
3D Printed Parts	~$500	Grippers, mounts, etc.
Other Hardware	~$2,840	Brackets, cables, etc.
Total	~$32,000	Officially stated on project page

Comparison: Existing commercial mobile manipulators (e.g., Clearpath + dual arms) are $200,000+

Physical Specifications

Source: Mobile ALOHA Project Page

Item	Spec
Footprint	90 cm x 135 cm
Arm Reach Height	65 cm ~ 200 cm
Arm Forward Extension	100 cm (from base)
Total Weight	75 kg
Pull Force	100 N @ 1.5 m height
Movement Speed	Up to 1.6 m/s

Differences from Static ALOHA

Item	ALOHA (Static)	Mobile ALOHA
Base	Fixed table	AgileX Tracer (mobile)
Action Dimensions	14-DoF (arms+grippers)	16-DoF (arms+grippers + base velocity)
Task Range	Tabletop manipulation	Full indoor environment
Teleop Method	Hand-operate leader arms	Whole-body teleop (walk while manipulating)
Cost	~$20,000	~$32,000
Load	Fixed	Self-balancing (using battery weight)

Action Space Expansion

DoF Explanation: Each ViperX 300 arm is 6-DoF (arm) + 1-DoF (gripper) = 7-DoF

ALOHA: 14-DoF joint positions
       [arm1(6) + gripper1(1) + arm2(6) + gripper2(1)]

Mobile ALOHA: 16-DoF
       [arm1(6) + gripper1(1) + arm2(6) + gripper2(1) + base_linear_vel(1) + base_angular_vel(1)]

This design allows existing imitation learning algorithms to be applied with minimal modification.

Co-Training: Core Technique

Motivation

Mobile bimanual manipulation datasets are sparse, but static bimanual manipulation data is abundant. Co-training improves performance by training these two types of data together.

Method

Training Data = Mobile ALOHA demos (50) + Static ALOHA datasets (existing)

Mobile data: Full 16-DoF actions
Static data: 14-DoF actions (base velocity padded with 0)

Effect

Condition	Average Success Rate
Mobile data only	~50%
With co-training	~84%
Improvement	+34%p

Demonstrated Tasks

Source: arXiv:2401.02117 Table 1

Success Rates (50 demos, with co-training)

Task	Success Rate	Description
Wipe Wine	100%	Wipe wine spill
Call Elevator	95%	Call elevator and board
Use Cabinet	85%	Open wall cabinet and store pot
High Five	85%	High five (20 demos)
Rinse Pan	80%	Rinse pan at kitchen sink
Push Chairs	80%	Organize chairs
Cook Shrimp	40%	Stir-fry shrimp (75 sec, only 20 demos used)

Task Categories

Cooking

Stir-fry and serve shrimp
Handle pots/pans
Rinse at sink

Cleaning/Organizing

Wipe wine spill
Push chairs to organize
Store items in cabinet
Use vacuum cleaner

Navigation + Manipulation

Press elevator button and board
Transport items between rooms

Interaction

High five
Hand items to people

Technical Details

Supported Algorithms

Algorithm	Description
ACT	Action Chunking Transformer
Diffusion Policy	Diffusion-based action generation
VINN	Visual Imitation through Nearest Neighbors

Simulation Environments

Transfer Cube
Bimanual Insertion

Training Settings

Item	Value
Number of Demos	50/task
Control Frequency	50 Hz
Image Resolution	640 x 480
Number of Cameras	3 (2 wrist + 1 front)

Open-Source Resources

Public Materials

Resource	Link
Paper	arXiv:2401.02117
Project Page	mobile-aloha.github.io
GitHub (Hardware)	mobile-aloha
GitHub (Algorithms)	act-plus-plus
Assembly Tutorial	Included in project page
3D Printing Files	Included in GitHub

Tutorial Contents

3D printing guide
Assembly sequence
Software installation
Calibration methods
Teleoperation usage

Research Team and Support

Authors

Name	Role
Zipeng Fu	Co-first author
Tony Z. Zhao	Co-first author
Chelsea Finn	Advisor

Support

Stanford Robotics Center
Steve Cousins
Stanford IRIS Lab members

Subsequent Developments

ALOHA 2 (Google DeepMind, Stanford, Hoku Labs, 2024)

Improved version jointly developed by Google DeepMind, Stanford, and Hoku Labs:

Gripper Operation Force Reduction: Leader arm gripper force requirement 14.68N to 0.84N (10x reduction)
2x Gripper Strength: Follower gripper max gripping force 12.8N to 27.9N
Material Upgrade: PLA/acrylic to 3D-printed carbon fiber nylon
Passive Gravity Compensation: Rubber bands to adjustable hanging retractors, 42% efficiency improvement
Large-Scale Data Collection: Capable of collecting 100s-1000s demonstrations per robot per day

Commercialization

Trossen Robotics sells ALOHA kits:

ALOHA Solo
ALOHA Bimanual Kit
Mobile ALOHA compatible parts

Significance and Impact

Academic Impact

Co-training Effect Proven: Performance improvement possible with related task data
Low-cost Research Platform: High-quality research possible at $32K
Reproducibility: Complete open-source enables replication in labs worldwide

Industrial Implications

“Mobile ALOHA has demonstrated something unique: relatively cheap robot hardware can solve really complex problems.” - Lerrel Pinto, Associate Professor of Computer Science, NYU

Demonstrated feasibility of household robots
Complex tasks possible even with low-cost hardware
Dramatically reduced data collection costs

References

Paper

@article{fu2024mobile,
  author    = {Fu, Zipeng and Zhao, Tony Z. and Finn, Chelsea},
  title     = {Mobile ALOHA: Learning Bimanual Mobile Manipulation
               with Low-Cost Whole-Body Teleoperation},
  journal   = {arXiv preprint arXiv:2401.02117},
  year      = {2024},
  note      = {Presented at CoRL 2024}
}