A Rapid Adapting and Continual Learning Spiking Neural Network Path Planning Algorithm for Mobile Robots

TL;DR

This work tackles online, cost-aware navigation for mobile robots by jointly learning traversal costs and planning paths. It introduces a Spiking Wavefront Planner (SWP) integrated with E-prop learning to iteratively adjust inter-neuron delays $D_{ij}$ based on observed costs, enabling rapid adaptation to changing environments. The system computes multi-cost maps (energy, obstacles, and slope) and demonstrates shorter, lower-cost paths than A* and RRT* in both real-world outdoor trials and large-scale simulations, with potential compatibility for neuromorphic hardware. While effective, the approach currently lacks generalization across unseen waypoints/environments and is computationally slower than some classic planners; future work targets vision-based cost estimation, memory replay, and hardware adaptations to enhance scalability and efficiency.

Abstract

Mapping traversal costs in an environment and planning paths based on this map are important for autonomous navigation. We present a neurobotic navigation system that utilizes a Spiking Neural Network Wavefront Planner and E-prop learning to concurrently map and plan paths in a large and complex environment. We incorporate a novel method for mapping which, when combined with the Spiking Wavefront Planner, allows for adaptive planning by selectively considering any combination of costs. The system is tested on a mobile robot platform in an outdoor environment with obstacles and varying terrain. Results indicate that the system is capable of discerning features in the environment using three measures of cost, (1) energy expenditure by the wheels, (2) time spent in the presence of obstacles, and (3) terrain slope. In just twelve hours of online training, E-prop learns and incorporates traversal costs into the path planning maps by updating the delays in the Spiking Wavefront Planner. On simulated paths, the Spiking Wavefront Planner plans significantly shorter and lower cost paths than A* and RRT*. The spiking wavefront planner is compatible with neuromorphic hardware and could be used for applications requiring low size, weight, and power.

Figures (7)

• Figure 1: Overview of our navigation system. 1. Neurons in the spiking wavefront planner represent locations in space. To plan a path, a spike is induced at the robot's start location and propagates until it reaches the goal location. 2. The robot navigates to the goal using the planned path. Internal measurements and environmental sensors track costs when traversing between neuron waypoints. 3. Using the learned costs, the planner is updated using E-prop. The figure shows each neuron's eligibility trace, which determines the update magnitude based on how recently the neuron spiked.
• Figure 2: Top down view of the Aldrich Park environment. Imagery © Google
• Figure 3: The Clearpath Jackal robot.
• Figure 4: The costmap for all costs added and normalized after learning. Nodes are colored according to the mean of the delays $D$ from other nodes. Example paths minimizing current drawn, obstacles encountered, and steepness are colored green, red, and blue, respectively.
• Figure 5: Mean squared error between the delays $D$ of the model at each training step and the final learned costs of the model.