Complex Obstacle Avoidance and Fleet Collision Avoidance: Gaussian Process Distance Field (GPDF) and Dynamic Window Approach (DWA)

For safe and flexible navigation in multi-robot systems, this work presents an enhanced and predictive sampling-based trajectory planning approach in complex environments, the Gradient Field-based Dynamic Window Approach (GF-DWA). Building upon the dynamic window approach, the proposed method utilizes gradient information of obstacle distances as a new cost term to anticipate potential collisions. This enhancement enables the robot to improve awareness of obstacles, including those with non-convex shapes.

The GPDF part is based on this work. The permission to use and extend the original GPDF code has been granted by its author.

Publication

The paper is accepted by IROS 2025.

Bibtex citation:

NOT YET

Quick Start

JAX

The GPDF is implemented in Google JAX. Please find the installation guide here.

Install dependencies

pip install -r requirements.txt

Use Case

There is no need to have any pre-training or pre-learned model. This method is off-the-shelf. Run main_base.py for the simulation (different scenarios and methods) in Python. The evaluation is activated by setting the evaluation variable to True. Change scenario_index and tracker_type in the code to select different scenarios and methods.

To watch the demo videos:

• Single-robot and multi-robot: Link

More videos from other projects are available on my personal page.