Viability-Preserving Passive Torque Control

TL;DR

The paper addresses safety lapses in traditional passivity-based torque control by introducing Viability-Preserving Passive Torque Control (VPP-TC), which pre-computes viable regions in augmented state space and enforces torque bounds via a QP. It fuses data-driven self-collision and external-collision viability with analytical joint-limit bounds, translating viability into torque constraints through the robot dynamics and a task-space impedance reference. Key contributions include a transformer-based self-collision viability classifier, Bernstein-polynomial SDFs for obstacle reasoning, and a unified QP-based controller that activates constraints adaptively; experiments on a 7-DoF Franka Panda show higher update rates and smoother trajectories than CPIC. This approach enables high-rate, safe, passive manipulation and teleoperation without requiring precomputed collision-free trajectories, with broad implications for human-robot interaction and safety-critical manipulation.

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 dynamics. 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.

Figures (7)

• Figure 1: Schematic of the proposed viability-preserving, passivity-based torque controller. Joint-limit torque bounds are derived analytically, whereas bounds for self-collision and external-object collisions are data-driven.
• Figure 2: Real-world teleoperation task - pressing a desk lamp button to turn it on. The human operator provides only coarse target positions (no precomputed collision-free trajectory). Our proposed torque controller VPP-TC ensures safety near the lamp and prevents self-collision while accomplishing the task.
• Figure 3: Viability concept: $s_1$ is feasible and viable, whereas $s_2$ is feasible but non-viable.
• Figure 4: Planar 3-DoF demo. Left: workspace trajectory. Right: viable box of $(\ddot q_0,\ddot q_1)$ under hardware limits $[-10,10]\ \mathrm{rad/s^2}$. With ECA enabled, the admissible box contracts as the obstacle constraint activates, illustrating viability-induced acceleration bounds.
• Figure 5: Braking rollout from $(q_0,\dot q_0)$ to rest $(q_e,\mathbf{0})$: label Safe iff every state along the trajectory to $(q_e,\mathbf{0})$ is self-collision-free; otherwise Self-Collide at the first contact $(q_t,\dot q_t)$. The resulting dataset trains a network that maps $(q,\dot q)$ to a viability score $\Gamma$.

Theorems & Definitions (3)

• Definition 1: Viability
• Definition 2: Self-collision-free viability
• Definition 3: External-collision-free viability