Sampling-free obstacle gradients and reactive planning in Neural Radiance Fields (NeRF)
This work addresses motion planning for robotics or autonomous systems in implicit neural representations, but it is incremental as it builds on existing NeRF and RMP methods.
The paper tackled the problem of motion planning in Neural Radiance Fields (NeRF) by augmenting a pre-trained NeRF to infer occupancy and approximate a Euclidean Signed Distance Field, enabling very fast sampling-free obstacle avoidance planning using backward differentiation and the Riemannian Motion Policies framework.
This work investigates the use of Neural implicit representations, specifically Neural Radiance Fields (NeRF), for geometrical queries and motion planning. We show that by adding the capacity to infer occupancy in a radius to a pre-trained NeRF, we are effectively learning an approximation to a Euclidean Signed Distance Field (ESDF). Using backward differentiation of the augmented network, we obtain an obstacle gradient that is integrated into an obstacle avoidance policy based on the Riemannian Motion Policies (RMP) framework. Thus, our findings allow for very fast sampling-free obstacle avoidance planning in the implicit representation.