Transformer-based Heuristic for Advanced Air Mobility Planning

TL;DR

A custom learning-based heuristic using transformer-based neural networks is developed, which significantly reduced computational load and enhanced the performance of the ASD A* algorithm.

Abstract

Safety is extremely important for urban flights of autonomous Unmanned Aerial Vehicles (UAVs). Risk-aware path planning is one of the most effective methods to guarantee the safety of UAVs. This type of planning can be represented as a Constrained Shortest Path (CSP) problem, which seeks to find the shortest route that meets a predefined safety constraint. Solving CSP problems is NP-hard, presenting significant computational challenges. Although traditional methods can accurately solve CSP problems, they tend to be very slow. Previously, we introduced an additional safety dimension to the traditional A* algorithm, known as ASD A*, to effectively handle Constrained Shortest Path (CSP) problems. Then, we developed a custom learning-based heuristic using transformer-based neural networks, which significantly reduced computational load and enhanced the performance of the ASD A* algorithm. In this paper, we expand our dataset to include more risk maps and tasks, improve the proposed model, and increase its performance. We also introduce a new heuristic strategy and a novel neural network, which enhance the overall effectiveness of our approach.

Figures (8)

• Figure 1: An example of heuristic generators: a) Riskmap2.0 takes the given start position (orange letter S), destination (green letter D), and risk map (black grid means the grid is blocked), generates heuristic (blue number) for all the available grid. b) Riskmap-state takes the current node (Blue letter A), destination, and risk map, generates heuristic (blue number) for the current node
• Figure 2: CSP example(2*2): The risk score of the grid (0, 0) is 0.1, the risk score of the grid (0, 1) and the grid (1, 1) is 0.05. There is a safety boundary(cyan plane) on the safety dimension, any path cross the safety boundary is invalid.
• Figure 3: Proposed transformer heuristic architecture
• Figure 4: Output of the Riskmap2.0(16*16)
• Figure 5: Concatenated embedding of the Riskmap-state(64*64)