EIZO inflates a line segment to a probabilistically collision-free polytope in robot configuration space. This polytope is guaranteed to contain the original line segment. It does this using a zero-order optimization approach that is massively parallelizable.

We propose a possible way of integrating EIZO into a motion planning pipeline that uses DRMs to find initial PWL paths. First, the DRM is constructed offline. Online, the DRM is then utilized to rapidly find a collision-free path, which is subsequently inflated and used by a decomposition-based motion planner to optimize a trajectory.  The optimized trajectory may be in collision since the sets are only probabilistically collision-free. To remedy this, we use the collisions found by the candidate trajectories to refine the sets until we find a collision-free trajectory.

We demonstrate the efficacy of our approach in simulation benchmarks in two and seven dimensions and validate the approach on hardware using a KUKA iiwa with perception in the loop. Relative to the nonlinear trajectory optimization baseline, our approach produces slightly higher-cost trajectories but increases the success rate from around 72% to 100% on our simulation benchmark while computing trajectories around 17 times faster.  The video demonstrates the hardware setup. For more details, please refer to the paper.