Safe navigation typically requires knowing the exact shapes of robots and obstacles, but irregular geometries make this hard. Control Barrier Functions (CBFs) are a common tool for enforcing safety, yet they often rely on simplified shapes like spheres or super-ellipsoids. Those surrogates are either too conservative in cluttered scenes or demand many local primitives, which bloats the number of constraints and slows down real-time control.

Researchers at the 2026 IEEE International Conference on Robotics and Automation propose a geometry-aware CBF built on Bernstein-Polynomial Signed Distance Fields (BP-SDFs). This representation lets the robot and obstacles be described in a unified way, so the barrier function works with a single minimum distance. Because Bernstein polynomials are differentiable, the control constraints can be enforced in a closed loop without extra approximation steps.

The authors validated their approach in simulations of single-robot navigation and heterogeneous multi-robot collision avoidance across different environments. The results indicate that the method maintains safety while keeping the computation efficient enough for real-time use. The paper is accepted at ICRA 2026 and available on arXiv.