Research Analyzer
← Back ICRA 2026

ALOHA Lightning: Learning Fast and Precise Manipulation

John Hua Yao, Qi Wu, Yihuai Gao, Chelsea Finn, Zipeng Fu

PDF

AI summary

Key figure (auto-extracted from paper)
ALOHA Lightning enables robots to learn and execute complex bimanual manipulation tasks at near-human speeds with high success rates using only 50 demonstrations per task.
imitation learning kinesthetic teaching high-speed manipulation bimanual robotics test-time smoothing visual masking

Problem

Learned robotic policies typically execute far slower than humans due to a lack of interfaces for collecting high-speed demonstration data and the difficulty of training robust policies for fast, unforgiving motions.

Approach

The system uses kinesthetic teaching on a backdrivable bimanual robot to collect near-human-speed trajectories, paired with a learning pipeline that applies test-time action smoothing and real-time visual masking to align training and deployment domains.

Key results

  • Achieves over 80% success rate on folding shorts and bussing tables with just 50 demonstrations per task
  • Completes dynamic toss-and-catch tasks in under 2 seconds with 50% success rate
  • Reduces task execution time by over 50% compared to traditional teleoperation systems
  • Visual masking and test-time smoothing significantly outperform baseline methods in success rate and stability

Why it matters

Bridges the critical speed gap in robotic manipulation, making high-speed, reliable bimanual skills viable for practical deployment in human-centric environments.

Abstract

Learning from human demonstrations has enabled robots to acquire a wide range of manipulation skills, but learned policies typically execute far slower than ordinary humans. This speed gap is mainly due to lack of an interface for collecting demonstration data at high speed, and the difficulty in training policies that can robustly execute high-speed motions. In this paper, we present ALOHA Lightning, a system for learning fast and precise robotic manipulation. Our system uses kinesthetic teaching to intuitively collect near-human-speed demonstrations on a backdrivable bimanual platform, yielding natural and fast trajectories. We also present a learning pipeline that enables smooth high-speed execution through test-time action smoothing and aligns the visual data distribution between data collection and deployment with masking. Given 50 demonstrations for each task, ALOHA Lightning autonomously completes bimanual tasks such as folding shorts, and bussing tables for over 80% success rates, and ball tossing and catching with 50% success rate while close to human speed.

Index terms

Imitation Learning Bimanual Manipulation

Related papers