New method evaluates safety for autonomous systems in moving environments

Geometry-Evolving Barrier Functions provide a way to track safety boundaries with lower computational costs.

Researchers introduced Geometry-Evolving Barrier Functions (GEBFs) for safety evaluation. These functions work in dynamic environments with multiple interacting agents.

The method propagates local safety geometry of a terminal-value Hamilton-Jacobi (HJ) partial differential equation (PDE). It evolves value, gradient, and curvature information along a Hamiltonian-induced reference trajectory.

Hamilton-Jacobi (HJ) reachability provides rigorous finite-horizon safety guarantees. A Hamilton-Jacobi (HJ) partial differential equation (PDE) is a mathematical tool used to solve for safety.

The GEBFs provide a compact representation of the local HJ safety landscape. An HJ-optimal characteristic is a path used to establish consistency between the GEBF and the HJ solution.

An iterative algorithm couples local geometric propagation with trajectory refinement. This approach achieves substantially lower computational cost than conventional HJ reachability.

Experiments on agent-obstacle collision-avoidance benchmarks show GEBFs accurately capture the evolution of dynamic safety boundaries. These tests included low- and high-dimensional systems.

Key facts

  • GEBFs provide a compact representation of the local HJ safety landscape.
  • The method achieves substantially lower computational cost than conventional HJ reachability.
  • GEBFs accurately capture the evolution of dynamic safety boundaries in collision-avoidance benchmarks.

All sections

Edition