Tony Zhao

ACT & ALOHA Lead Author, Sunday Robotics Co-founder/CEO

Profile

FieldDetails
Current PositionSunday Robotics Co-founder/CEO
PreviousStanford University PhD program (dropped out)
AdvisorChelsea Finn (IRIS Lab)
UndergraduateUC Berkeley EECS (2021)
Undergrad AdvisorsSergey Levine, Dan Klein

Key Contributions

  • ACT Lead Author: Action Chunking with Transformers, precise manipulation from 10-minute demonstrations
  • ALOHA Design: $20K low-cost bimanual robot system
  • Mobile ALOHA: Mobile bimanual robot
  • Sunday Robotics: Home robot Memo development
  • Open-Source Contributions: Full code and hardware design release

Research Timeline

UC Berkeley (2017-2021)

Advised by Sergey Levine, Dan Klein

YearWorkImpact
2021EECS Bachelor’sAI/ML research foundation

Stanford PhD Program (2021-2024)

Advised by Chelsea Finn (IRIS Lab)

YearWorkImpact
2021PhD StartedIRIS Lab, Stanford Robotics Fellowship (2022-23)
2023ACTAction Chunking, learning from 10-minute demonstrations
2023ALOHA$20K bimanual system (RSS 2023)
2024Mobile ALOHAMobile bimanual robot (CoRL 2024)
2024Dropped out of PhDFounded Sunday Robotics

Sunday Robotics (2024-present)

YearWorkImpact
2024Co-founded Sunday RoboticsCEO alongside Cheng Chi (CTO)
2025Memo launchHome robot, $35M funding raised

Major Publications

ACT & ALOHA (RSS 2023)

“Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware”

Key contributions:

  • ACT Algorithm: Action Chunking + CVAE
  • ALOHA Hardware: $20K low-cost bimanual system
  • 80-90% success rate with 50 demonstrations

Mobile ALOHA (2024)

“Mobile ALOHA: Learning Bimanual Mobile Manipulation with Low-Cost Whole-Body Teleoperation”

Key contributions:

  • Mobile platform + ALOHA combination
  • Whole-body teleoperation
  • Household task execution

Key Ideas

Action Chunking (2023)

Core: Predict action sequences (chunks) instead of single actions

Previous: Observation -> Next 1 action
ACT: Observation -> Next k action sequence

Advantages:
- Mitigates compounding error
- Effective horizon reduced by k times
- Smooth motion generation

ALOHA Hardware

Components:
- 2x ViperX 6-DoF arms (~$5,600 each)
- 4x RGB cameras
- Teleoperation rig

Features:
- Total cost ~$20K (1/10 of typical)
- Modular design
- Dynamixel motors (easy replacement)
- 1.5m working range

Impact

Research Democratization

  • Low-Cost: $20K enables lab reproducibility
  • Open-Source: Full code and hardware design released
  • LeRobot Integration: HuggingFace default model

Follow-up Research

  • Mobile ALOHA
  • ALOHA 2 (Google)
  • Numerous ALOHA-based research projects

Philosophy

Research Philosophy

“Make complex things simple. Make expensive things affordable. Make closed things open.”

Open-Source Contributions

  • Full ACT code release
  • ALOHA hardware design release
  • Training data release
  • Detailed documentation

See Also