An ablation study evaluating the role of heuristic diversity in Multi-Heuristic A* (MHA*) within cost-aware, synthetic Mars-like terrain maps.
This project deconstructs the MHA* algorithm to determine if its performance gains stem from its multi-queue architecture or the diversity of its heuristics.
- Planners Implemented: Weighted A* (WA*), MHA* with Diverse Heuristics, and MHA* with Homogeneous Heuristics.
- Heuristics Used: Euclidean (Anchor), Scaled Manhattan, and Aggressive Euclidean.
- Custom Terrain Model: Synthetic 100x100 grid maps with varying traversal costs for Bedrock (1), Gravel (5), and Sand (10).
Figure 1: Comparison of paths generated by different planners across Mars-like terrain.
Experiments across 500 scenarios show that MHA* with diverse heuristics achieves orders-of-magnitude reductions in node expansions and planning time compared to WA* in complex terrains.
Figure 2: Performance metrics highlighting the efficiency of diverse heuristics.
- Python 3.x
- NumPy
- Matplotlib
- Seaborn
- SciPy
Run the mars_navigation_planner.ipynb notebook to:
- Generate the terrain data.
- Execute the path-planning experiments.
- Visualize the performance metrics.