Control Barrier Functions for Safety-Aware Latent-Space Motion Planning
Control Barrier Functions for Safety-Aware Latent-Space Motion Planning
This research proposes replacing binary collision classifiers with neural control barrier functions (CBFs) in latent-space path planning for robotic manipulators. By training a CBF with safe/unsafe sign constraints and discrete-time decrease conditions, the method enforces safety via first-order corrections during planning. On a 7-DoF Franka Panda, it achieves a 75.6% success rate on fixed-obstacle tasks (12.0 percentage points above the best of 35 classifier configurations) and 83.8% on multi-position tasks. A decay rate sweep identifies optimal safety-performance trade-offs in α ∈ [0.10, 0.20], though random-obstacle settings yield a 33.0% success rate due to covariate shift between obstacle-blind training and corrected planner states.
The study introduces a neural control barrier function (CBF) within latent-space motion planning to address gradient saturation and lack of temporal structure in existing binary classifier approaches, achieving a 75.6% success rate on fixed obstacles with a 7-DoF Franka Panda manipulator while attributing performance drops in random-obstacle settings to covariate shift.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
decay rate alpha range
0.1
View original evidence
characterizes the safety--performance trade-off, with the best operating points in $\alpha \in [0.10, 0.20]$. In a fully general random-obstacle setting, success drops to 33.0\%, which we trace to covariate shift between the obstacle-blind training transitions and the states visited by the corrected planner.
Open source S5decay rate alpha range upper bound
0.2
View original evidence
characterizes the safety--performance trade-off, with the best operating points in $\alpha \in [0.10, 0.20]$. In a fully general random-obstacle setting, success drops to 33.0\%, which we trace to covariate shift between the obstacle-blind training transitions and the states visited by the corrected planner.
Open source S5robot degrees of freedom
7
View original evidence
planner, and it is enforced at planning time by a closed-form, first-order correction that projects each goal-directed latent step back onto the CBF constraint. On a 7-DoF Franka Panda manipulator, the CBF planner achieves a 75.6\% success rate on a fixed-obstacle task, 12.0 percentage points above the best of 35 tuned classifier configurations, and 83.8\% on a multi-position task. A sweep of the decay rate $\alpha$
Open source S4
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
planner, and it is enforced at planning time by a closed-form, first-order correction that projects each goal-directed latent step back onto the CBF constraint. On a 7-DoF Franka Panda manipulator, the CBF planner achieves a 75.6\% success rate on a fixed-obstacle task, 12.0 percentage points above the best of 35 tuned classifier configurations, and 83.8\% on a multi-position task. A sweep of the decay rate $\alpha$
Open source S4
characterizes the safety--performance trade-off, with the best operating points in $\alpha \in [0.10, 0.20]$. In a fully general random-obstacle setting, success drops to 33.0\%, which we trace to covariate shift between the obstacle-blind training transitions and the states visited by the corrected planner.
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org