aiqbench analysis
This paper addresses a real control problem: learned whole-body policies at 50 Hz are too slow to handle the first impact peak, so the authors add a non-learned tactile reflex pathway that drives motors within 10 ms. The reported gains (34.1% lower first force peak, 2.15x lower contact impulse versus a 20 ms-delayed reflex) come from a single robot platform with a small set of contact scenarios, so they need replication before generalising. The most useful takeaway for AI readers is the hybrid design: fast deterministic reflexes gate and constrain learned policies, which may matter more than the learning method itself.
ReflexArc: Bottom-Up Integration of Instantaneous Tactile Reflexes into Whole-Body Humanoid Control
arXiv:2611.04424v1 Announce Type: new Abstract: Humanoid robots operating in cluttered environments must react to unexpected contacts before the first impact peak, which arrives within tens of milliseconds. Learned whole-body policies running at 50 Hz introduce a policy-cycle delay and may suppress a protective withdrawal as a tracking error. To address these challenges, we propose ARC (Ascending Reflex Control), a bottom-up framework that integrates instantaneous tactile reflexes with learned whole-body control. ARC consists of two components: an instantaneous reflex pathway and bottom-up reflex integration. The reflex pathway detects hazardous contacts from full-body tactile sensing, solves a whole-body withdrawal pose, and delivers it directly to the motors within 10 ms. Bottom-up reflex integration combines reflex-in-the-loop policy learning, reflex-transparent policy alignment, and reflex-informed policy adaptation to sustain the withdrawal, maintain balance, and complete longer-horizon recovery. We deploy ARC on a Unitree G1 and evaluate three categories of hazardous contact. Compared with an otherwise identical reflex delayed by 20 ms, ARC reduces the first force peak by 34.1% and reduces the contact impulse from 0.2632 N s to 0.1225 N s (a 2.15x ratio). We further demonstrate protected motion tracking, teleoperation, and vision-free obstacle traversal guided by foot tactile sensing. Project page: https://minusmodulo.github.io/reflexarc-website/
Abstract:Humanoid robots operating in cluttered environments must react to unexpected contacts before the first impact peak, which arrives within tens of milliseconds. Learned whole-body policies running at 50 Hz introduce a policy-cycle delay and may suppress a protective withdrawal as a tracking error. To address these challenges, we propose ARC (Ascending Reflex Control), a bottom-up framework that integrates instantaneous tactile reflexes with learned whole-body control. ARC consists of two components: an instantaneous reflex pathway and bottom-up reflex integration. The reflex pathway detects hazardous contacts from full-body tactile sensing, solves a whole-body withdrawal pose, and delivers it directly to the motors within 10 ms. Bottom-up reflex integration combines reflex-in-the-loop policy learning, reflex-transparent policy alignment, and reflex-informed policy adaptation to sustain the withdrawal, maintain balance, and complete longer-horizon recovery. We deploy ARC on a Unitree G1 and evaluate three categories of hazardous contact. Compared with an otherwise identical reflex delayed by 20 ms, ARC reduces the first force peak by 34.1% and reduces the contact impulse from 0.2632 N s to 0.1225 N s (a 2.15x ratio). We further demonstrate protected motion tracking, teleoperation, and vision-free obstacle traversal guided by foot tactile sensing. Project page: this https URL
Robotics (cs.RO)
arXiv:2611.04424 [cs.RO]
(or arXiv:2611.04424v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2611.04424 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 289 words
Related coverage
Emergent Underactuation in Fully Actuated Multirotors by Minimizing a Homogeneous Allocation Cost
arXiv · Oct 6, 2026
NORi: An ML-Augmented Ocean Boundary Layer Parameterization
arXiv · Aug 5, 2026
Mitigating Compiler Fusion-Induced Power Bursts in Mobile NPU Inference as the Battery Depletes
arXiv · Jul 21, 2026
Trust Base partners with OQC to explore quantum finance workflows
Quantum Zeitgeist · Sep 16, 2026