Safe, Task-Consistent Manipulation with Operational Space Control Barrier Functions

TL;DR

This work presents Operational Space Control Barrier Functions (OSCBF) as a real-time, safety-filter framework that preserves task-consistent behavior while enforcing hundreds of safety constraints in both operational and joint spaces. By combining CBF theory with hierarchical task prioritization, the approach supports both velocity- and torque-controlled manipulators, scales to large constraint sets via a high-performance QP solver, and maintains real-time rates on hardware. The method is validated in simulation and on a Franka Panda, demonstrating singularity avoidance, collision avoidance, and workspace containment in highly cluttered and dynamic scenarios, with open-source software available. The OSCBF framework thus enables safe, scalable manipulation suitable for learning-based control and teleoperation in unstructured environments.

Abstract

Safe real-time control of robotic manipulators in unstructured environments requires handling numerous safety constraints without compromising task performance. Traditional approaches, such as artificial potential fields (APFs), suffer from local minima, oscillations, and limited scalability, while model predictive control (MPC) can be computationally expensive. Control barrier functions (CBFs) offer a promising alternative due to their high level of robustness and low computational cost, but these safety filters must be carefully designed to avoid significant reductions in the overall performance of the manipulator. In this work, we introduce an Operational Space Control Barrier Function (OSCBF) framework that integrates safety constraints while preserving task-consistent behavior. Our approach scales to hundreds of simultaneous constraints while retaining real-time control rates, ensuring collision avoidance, singularity prevention, and workspace containment even in highly cluttered settings or during dynamic motions. By explicitly accounting for the task hierarchy in the CBF objective, we prevent degraded performance across both joint-space and operational-space tasks, when at the limit of safety. We validate performance in both simulation and hardware, and release our open-source high-performance code and media on our project webpage, https://stanfordasl.github.io/oscbf/

Figures (6)

• Figure 1: Safe and performant highly-constrained manipulation. Our OSCBF controller maintains safety for hundreds of constraints enforced concurrently, using the full second-order robot dynamics and torque control, and operates at real-time control rates (over 1000 Hz). Shown above is the simultaneous constraint evolution for all CBFs during one adversarial trajectory from teleoperation (end-effector target shown in red). Across 168 safety conditions, the robot remains safe ($h(\mathbf{z}) >0$) without over-conservative behavior near the boundary of safety ($h(\mathbf{z}) =0$).
• Figure 2: Task-consistency: balancing safety and performance. Consider the behavior of the robot with the tip of the end-effector at the boundary of safety, where the desired goal moves towards the unsafe region. If the CBF objective is not task-consistent, the safety filter leads to a decrease in performance, even if safety is maintained. In (A), a joint space metric leads to a decrease in the operational space task performance, and in (B), a purely operational space metric leads to excess motion in the null space. (C) With OSCBF, excess motion is minimized and both safety and task performance is maintained.
• Figure 3: Block diagram: OSCBF for torque-controlled manipulators
• Figure 4: Scaling up collision avoidance. Even in highly-cluttered scenes, our OSCBF controller maintains safety, task-tracking performance, and real-time control rates. Consider a tabletop environment (inset) with many randomly-generated collision bodies, shown in blue. OSCBF scales to over 400 CBF constraints while retaining real-time control rates (1000 Hz) for torque control, and well over 1000 constraints while retaining good control rates (100 Hz) for velocity control. We indicate both the mean frequency and the minimum frequency, assuming an allowable 5% packet drop.
• Figure 5: Dynamic task consistency under input constraints. Under high-speed unsafe motions, accounting for the full dynamics of the robot and torque input constraints is necessary for good task performance. Consider a periodic, straight-line end-effector trajectory that commands a rapid motion of the end-effector tip into the unsafe set. With constraints on both the maximum joint velocities and torque, both a velocity-controlled robot (left) and a torque-controlled robot (right) maintain safety. However, the velocity CBF causes a degradation of tracking performance, due to instantaneously infeasible velocity commands, given the configuration of the robot and its torque limits.