FACTR 2: Learning External Force Sensing for Commodity Robot Arms Improves Policy Learning

2026-06-10 · Source: Artificial Intelligence · Field: Technology & Digital — Robotics & Autonomous Systems, Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

FACTR 2 introduces a novel approach to enable force sensitivity in commodity robot arms lacking dedicated force sensors. This system comprises Neural External Torque Estimation (NEXT) and Force-Informed Re-Sampling Training (FIRST). NEXT is a data-driven method estimating external joint torques. It trains in just 1 minute from 10 minutes of free-motion data, achieving estimates comparable to dedicated joint-torque sensors. This capability facilitates force-feedback teleoperation on low-cost robotic platforms. Complementing NEXT, FIRST enhances policy learning by up-sampling pre-contact and contact segments during behavior cloning. Across five long-horizon tasks, FIRST demonstrated a performance improvement of over 17% in task progress compared to previous force-aware policies. Together, FACTR 2's components bring advanced force-aware teleoperation and policy learning to off-the-shelf robots without requiring additional sensing hardware.

Key takeaway

For Robotics Engineers developing contact-rich manipulation tasks on commodity robot arms, FACTR 2 offers a significant pathway to advanced capabilities. You can implement Neural External Torque Estimation (NEXT) to gain force-feedback for teleoperation. Apply Force-Informed Re-Sampling Training (FIRST) to boost policy learning performance by over 17%. This allows your team to achieve sophisticated force-aware behaviors using existing off-the-shelf robotic systems, accelerating development and reducing hardware expenditure.

Key insights

FACTR 2 enables force-aware robotics on commodity arms by estimating torques and optimizing policy learning without extra sensors.

Principles

• Data-driven torque estimation can match dedicated sensors.
• Targeted data re-sampling improves contact-rich policy learning.
• Low-cost hardware can achieve advanced force-feedback.

Method

NEXT trains in 1 minute from 10 minutes of free-motion data to estimate external joint torques. FIRST up-samples pre-contact and contact segments during behavior cloning for policy learning.

In practice

• Enable force-feedback teleoperation on low-cost arms.
• Improve policy learning for contact-rich manipulation.
• Utilize existing robot hardware for advanced sensing.

Topics

• Robot Force Sensing
• Commodity Robot Arms
• Neural External Torque Estimation
• Policy Learning
• Behavior Cloning
• Robotics Teleoperation

Best for: Research Scientist, Robotics Engineer, Machine Learning Engineer, AI Scientist

Related on AIssential

• TAM: Torque Adaptation Module for Robust Motion Transfer in Manipulation — TAM enables robust robot policy transfer by adapting torques using proprioceptive history, trained purely in simulation.
• TouchThinker: Scaling Tactile Commonsense Reasoning to the Open World with Large-scale Data and Action-aware Representation — TouchThinker scales open-world tactile commonsense reasoning through large-scale data and action-aware representation.
• AI-powered robot beats elite table tennis players — Sony AI's Ace robot demonstrates advanced real-time perception and control in complex human-robot interaction.
• Dense Force Estimation with an Event-based Optical Tactile Sensor — Event-based tactile sensors can reconstruct dense 3D force fields by mapping displacement data to forces via iFEM and neural networks.
• How can robots acquire skills through interactions with the physical world? An interview with Jiaheng Hu — SLAC enables safe and sample-efficient real-world reinforcement learning for complex robots by pretraining a latent action space in low-fidelity simulation.
• Training a Robotic Arm Using MuJoCo and JAX on AMD Hardware with ROCm™ — Reinforcement learning with MuJoCo and JAX on AMD ROCm enables robust robotic arm training for complex manipulation tasks.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.