Long Short-Term Memory for Spatial Encoding in Multi-Agent Path Planning

What is it about?

Multi-agent drone systems constitute a research topic of high interest with progress being made in the field of unmanned aerial vehicles. One key aspect for these autonomous systems is the perception of their surrounding. Clever encoding of the massive amount of data produced by sensors is crucial for the success of autonomous systems. This paper presents such a method using Long Short-Term Memory, a type of recurrent neural network. The encoding is coupled with deep learning-based path planning for collision avoidance and arrival at a target position. Furthermore, a framework for interpreting the learned encoding behavior is included. The paper concludes with real-world flight tests that demonstrate the feasibility of the theoretically developed method.

Featured Image