A Safety-Aware Shared Autonomy Framework with BarrierIK Using Control Barrier Functions

Berk Guler; Kay Pompetzki; Yuanzheng Sun; Simon Manschitz; Jan Peters

cs.ROMar 2, 2026

A Safety-Aware Shared Autonomy Framework with BarrierIK Using Control Barrier Functions

Authors: Berk Guler, Kay Pompetzki, Yuanzheng Sun, Simon Manschitz, Jan Peters

Organizations: Technical University of Darmstadt · Honda Research Institute Europe GmbH · German Research Center for Artificial Intelligence (DFKI) · Robotics Institute Germany (RIG) · Centre for Cognitive Science

Abstract

Shared autonomy blends operator intent with autonomous assistance. In cluttered environments, linear blending can produce unsafe commands even when each source is individually collision-free. Many existing approaches model obstacle avoidance through potentials or cost terms, which only enforce safety as a soft constraint. In contrast, safety-critical control requires hard guarantees. We investigate the use of control barrier functions (CBFs) at the inverse kinematics (IK) layer of shared autonomy, targeting post-blend safety while preserving task performance. Our approach is evaluated in simulation on representative cluttered environments and in a VR teleoperation study comparing pure teleoperation with shared autonomy. Across conditions, employing CBFs at the IK layer reduces violation time and increases minimum clearance while maintaining task performance. In the user study, participants reported higher perceived safety and trust, lower interference, and an overall preference for shared autonomy with our safety filter. Additional materials available at https://berkguler.github.io/barrierik.

Figures & tables

Fig. 1 : Pipeline with CBF-based safety filtering after arbitration. The operator specifies a target pose $\mathbf{T}_{\mathrm{h}}(t)$ ; the guiding policy outputs $\mathbf{T}_{\mathrm{r}}(t)$ from the current state $\bm{s}(t)~=~\{\bm{\theta}(t),\bm{\mathcal{O}}(t),\bm{\mathcal{L}}(t),\text{task context}\}$ , where $\bm{\theta}$ is the joint configuration, $\bm{\mathcal{O}}$ and $\bm{\mathcal{L}}$ are obstacle and robot link capsules, and “task context” includes task-specific info (e.g., rendered views). These unfiltered poses are blended in $SE(3)$ via arbitration weight $\alpha(t)$ to form $\mathbf{T}(t)$ (red dashed), which may violate safety margins. BarrierIK receives $\mathbf{T}(t)$ and enforces CBF conditions to compute a safe joint command $\bm{\theta}^{*}(t)$ (green), which is applied to the robot. The updated state $\bm{s}(t)$ is fed back to the policy and user.

Fig. 2 : Simulated task and collision-evaluation setup. Left: scene with human target $\mathbf{T}_{\text{h}}(t)$ and autonomous reference $\mathbf{T}_{\text{r}}(t)$ . Right: robot link (red) and obstacle (black) convex-capsule colliders; for each link $\ell\in\mathcal{L}$ and obstacle $o\in\mathcal{O}$ we compute signed distances $\phi_{\ell o}(\bm{\theta})$ and signed distance vectors used by Baseline P and BarrierIK (colors encode magnitude).

Task	Solver	#Collisions	Min. clearance [m]	Violation time [%]	Pos. err. [m]	Ori. err. [ ∘ ]	Task jerk [m/s 3 ]	Joint jerk [rad/s 3 ]
Shelf	N	44.13 $\,\pm\,$ 9.74	-0.162 $\,\pm\,$ 0.016	24.52 $\,\pm\,$ 1.19	0.020 $\,\pm\,$ 0.001	4.15 $\,\pm\,$ 0.07	4.197 $\,\pm\,$ 0.354	48.15 $\,\pm\,$ 2.89
	P	19.30 $\,\pm\,$ 5.85	-0.171 $\,\pm\,$ 0.017	4.26 $\,\pm\,$ 5.22	0.049 $\,\pm\,$ 0.018	4.50 $\,\pm\,$ 0.12	6.518 $\,\pm\,$ 0.738	57.22 $\,\pm\,$ 3.35
	B	16.80 $\,\pm\,$ 4.31	-0.153 $\,\pm\,$ 0.020	0.53 $\,\pm\,$ 0.19	0.043 $\,\pm\,$ 0.009	4.76 $\,\pm\,$ 0.49	6.037 $\,\pm\,$ 0.496	65.89 $\,\pm\,$ 6.22
Dynamic	N	13.43 $\,\pm\,$ 3.30	-0.091 $\,\pm\,$ 0.025	56.25 $\,\pm\,$ 16.62	0.002 $\,\pm\,$ 0.004	7.26 $\,\pm\,$ 0.13	2.835 $\,\pm\,$ 0.437	45.55 $\,\pm\,$ 3.51
	P	1.55 $\,\pm\,$ 1.80	-0.033 $\,\pm\,$ 0.022	4.04 $\,\pm\,$ 9.17	0.023 $\,\pm\,$ 0.015	7.35 $\,\pm\,$ 0.25	3.219 $\,\pm\,$ 0.911	33.62 $\,\pm\,$ 6.25
	B	2.05 $\,\pm\,$ 3.74	-0.024 $\,\pm\,$ 0.022	0.84 $\,\pm\,$ 1.92	0.017 $\,\pm\,$ 0.010	8.13 $\,\pm\,$ 1.32	5.949 $\,\pm\,$ 1.927	54.47 $\,\pm\,$ 4.92

TABLE I : Autonomous trajectory evaluation (mean

\,\pm\,

s.d.). For Min. clearance , higher is better (negative

\Rightarrow

penetration). For all other metrics, lower is better. Bold indicates a statistically significant B vs. P difference (N serves only as a reference; excluded from the significance analysis test).

Fig. 3 : The stacked bar chart illustrates the number of participants who ranked each controller configuration from 1st to 6th choice. Each color segment within a bar represents the count of users selecting that configuration at the respective rank.

Fig. 4 : Subjective workload and usability across teleoperation configurations. (a) Polar (“radar”) plot of modified NASA-TLX dimensions—physical demand (I-PD), temporal demand (I-TD), mental demand (I-MD), performance (I-PER), effort (I-EFF), frustration (I-FL), control level (CL), assistance level (AL), and safety level (SL)—with “I-” denoting inverted scales so that larger values correspond to more favorable assessments. (b) Violin plots of participants’ aggregated NASA-TLX scores for each configuration (mean

\circ

, median

\blacklozenge

), where lower workload (higher support) appears toward the top of the scale.

Fig. 5 : Objective evaluation across six system configurations. Subplots (a) – (c) report performance and safety metrics: success rate (higher is better), task completion time (lower is better, measured only for successful trials), number of collisions (lower is better).

\blacklozenge

indicates the median, asterisks indicate statistical significance: ∗

p<.05

, ∗∗

p<.01

, ∗∗∗

p<.001

Explore similar work

Is Your Safe Controller Actually Safe? A Critical Review of CBF Tautologies and Hidden Assumptions
Mar 7, 2026Taekyung KimControl Barrier Functions Safety-Critical Control
Department of Robotics, University of Michigan, Ann Arbor, MI, 48109, USA
VertexCBF: Improving Neural Control Barrier Functions via Vertex-Restricted Control Search
Sep 11, 2026Bojan Derajić, Sebastian Bernhard, Wolfgang HönigControl Barrier Functions Safe Robot Navigation
Technical University of Berlin, Germany · AUMOVIO, Germany · Technical University of Applied Sciences Augsburg, Germany +1
BIG-CBF: Behavior-Imagination-Guided Control Barrier Function with Shared Uncertainty for Mobile Robot Navigation
Sep 13, 2026Shibo Li, Zhongcheng Wang, Jiahe Cao +2Robot Navigation Safe Robot Navigation
School of Automation, Northwestern Polytechnical University, Xi’an, China · Department of Robotics, Mohamed bin Zayed University of Artificial Intelligence, Abu Dhabi, United Arab Emirates

Mar 7, 2026J/K move · Enter open · S save

Is Your Safe Controller Actually Safe? A Critical Review of CBF Tautologies and Hidden Assumptions

Taekyung Kim

Department of Robotics, University of Michigan, Ann Arbor, MI, 48109, USA

Control Barrier Functions Safety-Critical Control Robot Safety Safe Robot Navigation

Open paper AskPDF arXiv

This tutorial provides a critical review of the practical application of Control Barrier Functions (CBFs) in robotic safety. While the theoretical foundations of CBFs are well-established, I identify a recurring gap between the mathematical assumption of a safe controller's existence and its constructive realization in systems with input constraints. I highlight the distinction between candidate and valid CBFs by analyzing the interplay of system dynamics, actuation limits, and class-K functions. I further show that some purported demonstrations of safe robot policies or controllers are limited to passively safe systems, such as single integrators or kinematic manipulators, where safety is already inherited from the underlying physics and even naive geometric hard constraints suffice to prevent collisions. By revisiting simple low-dimensional examples, I show when CBF formulations provide valid safety guarantees and when they fail due to common misuses. I then provide practical guidelines for constructing realizable safety arguments for systems without such passive safety. A crowd-navigation simulation study further illustrates that CBF-derived reward shaping in reinforcement learning can improve empirical behavior without establishing formal safety. The goal of this tutorial is to bridge the gap between theoretical guarantees and actual implementation, supported by an open-source interactive web demonstration that visualizes these concepts intuitively.

Abstract

Explore similar work

Is Your Safe Controller Actually Safe? A Critical Review of CBF Tautologies and Hidden Assumptions

VertexCBF: Improving Neural Control Barrier Functions via Vertex-Restricted Control Search

BIG-CBF: Behavior-Imagination-Guided Control Barrier Function with Shared Uncertainty for Mobile Robot Navigation