Robot navigation traditionally relies on building an explicit map that is used to plan collisionfree trajectories to a desired target. In deformable, complex terrain, using geometric-based approaches can fail to find a path due to mischaracterizing deformable objects as rigid and impassable. Instead, we learn to predict an estimate of traversability of terrain regions and to prefer regions that are easier to navigate (e.g., short grass over small shrubs). Rather than predicting collisions, we instead regress on realized error compared to a canonical dynamics model. We train with an on-policy approach, resulting in successful navigation policies using as little as 50 minutes of training data split across simulation and real world. Our learningbased navigation system is a sample efficient shortterm planner that we demonstrate on a Clearpath Husky navigating through a variety of terrain including grassland and forest.

No abstract

No abstract

Recent research has enabled fixed-wing unmanned aerial vehicles (UAVs) to maneuver in constrained spaces through the use of direct nonlinear model predictive control (NMPC) [1]. However, this approach has been limited to a priori known maps and ground truth state measurements. In this paper, we present a direct NMPC approach that leverages NanoMap [2], a light-weight point-cloud mapping framework to generate collision-free trajectories using onboard stereo vision. We first explore our approach in simulation and demonstrate that our algorithm is sufficient to enable vision-based navigation in urban environments. We then demonstrate our approach in hardware using a 42-inch fixed-wing UAV and show that our motion planning algorithm is capable of navigating around a building using a minimalistic set of goal-points. We also show that storing a point-cloud history is important for navigating these types of constrained environments.

No abstract