← Back ICRA 2026

Viability-Preserving Passive Torque Control

Zizhe Zhang, Yicong Wang, Zhiquan Zhang, Tianyu Li, Nadia Figueroa

PDF

AI summary

Key figure (auto-extracted from paper)

A viability-based framework pre-computes safe joint state sets to enforce collision and joint-limit constraints on passive torque controllers, enabling safer, smoother, and faster real-time robot control.

Viability theory Passive torque control Collision avoidance Quadratic programming Physical human-robot interaction Real-time safety

Problem

Conventional passivity-based torque controllers lack explicit constraint handling, making it difficult to guarantee collision avoidance and joint-limit satisfaction under external perturbations without causing optimization infeasibility or high computational overhead.

Approach

The method pre-computes safe joint position and velocity sets using viability theory and data-driven models, then enforces these as adaptive torque constraints in a quadratic program to maintain passivity while guaranteeing long-term safety.

Key results

Transformer-based self-collision viability classifier
Bernstein polynomial external obstacle distance estimator
Viability-induced torque constraints in a QP framework
Higher control-loop rates and smoother trajectories than baseline CPIC

Why it matters

It enables safe physical human-robot interaction by guaranteeing collision avoidance and closed-loop stability under external perturbations without sacrificing real-time control performance.

Abstract

Conventional passivity-based torque controllers for manipulators are typically unconstrained, which can lead to safety violations under external perturbations. In this paper, we employ viability theory to pre-compute safe sets in the state-space of joint positions and velocities. These viable sets, constructed via data-driven and analytical methods for self- collision avoidance, external object collision avoidance and joint-position and joint-velocity limits, provide constraints on joint accelerations and thus joint torques via the robot dy- namics. A quadratic programming-based control framework enforces these constraints on a passive controller tracking a dynamical system, ensuring the robot states remain within the safe set in an infinite time horizon. We validate the proposed approach through simulations and hardware experiments on a 7-DoF Franka Emika manipulator. In comparison to a baseline constrained passive controller, our method operates at higher control-loop rates and yields smoother trajectories. Project Website: vpp-tc.github.io

Index terms

Robot Safety Collision Avoidance Physical Human-Robot Interaction