DTG : Diffusion-based Trajectory Generation for Mapless Global Navigation

TL;DR

A novel Conditional RNN(CRNN) for diffusion models to efficiently generate trajectories to minimize the travel distance and maximize the traversability by choosing paths that do not lie in undesirable areas is presented.

Abstract

We present a novel end-to-end diffusion-based trajectory generation method, DTG, for mapless global navigation in challenging outdoor scenarios with occlusions and unstructured off-road features like grass, buildings, bushes, etc. Given a distant goal, our approach computes a trajectory that satisfies the following goals: (1) minimize the travel distance to the goal; (2) maximize the traversability by choosing paths that do not lie in undesirable areas. Specifically, we present a novel Conditional RNN(CRNN) for diffusion models to efficiently generate trajectories. Furthermore, we propose an adaptive training method that ensures that the diffusion model generates more traversable trajectories. We evaluate our methods in various outdoor scenes and compare the performance with other global navigation algorithms on a Husky robot. In practice, we observe at least a 15% improvement in traveling distance and around a 7% improvement in traversability.

Figures (8)

• Figure 1: DTG generates trajectories in traversable areas with the shortest travel distance to the target (red star). The blue and yellow boxes show the efficacy of the generated trajectories in scenarios including bushes and buildings.
• Figure 2: DTG Architecture: DTG has two models: Perception Encoder and Diffusion Model. Perception Encoder encodes the observation information, $\mathbf{o} = \left\{o_l, o_v , g\right\}$, to the condition vector, $\mathbf{c}$. The Diffusion Model takes a Gaussian distribution to generate a trajectory $\hat{\bm{\tau}}$ under the condition $\mathbf{c}$.
• Figure 3: Traversability Analysis in Challenging Occluded Environment: The top row shows the generated trajectories (red) in the camera view. The bottom row shows the top-down view of the traversability map. The cyan color represents the generated trajectories, and the yellow color represents the most heuristic trajectory to the goal. DTG can generate trajectories w.r.t. the geometric shape of the traversable areas, but other approaches cannot generate fully traversable trajectories; the non-traversable parts are marked by red circles.
• Figure 4: Travel-Distance Analysis in Challenging Narrow Passage Environment: The goal is behind the building. DTG generates a trajectory in a narrower space instead of the wide main road, leading to a shorter travel distance to the goal.
• Figure 5: The red star indicates the start position and the yellow circle indicates the goal. The example trajectory from the start to the goal is the blue path, with around a 200-meter travel distance.