Real-Time A* — Heuristic Search Visualizer
University AI course — 3-person team project (built together)
A real-time heuristic search engine: at each step the agent does a depth-limited lookahead, applies the RTA* learning rule to update its heuristic in memory, and commits to the best move — never planning the whole path up front. A three-person Artificial Intelligence course project, built collaboratively (the repository lives under a teammate’s account).
- Algorithm in SWI-Prolog: RTA* (Korf, 1990) on a configurable grid, Manhattan-distance heuristic, with a full A* for comparison; dynamic predicates store learned heuristic values.
- Python GUI via PySwip: a Tkinter canvas renders the live lookahead frontier with a per-step f-value colour gradient, plus timing and node-expansion stats.
- Variants explored: alpha-pruning to cut hopeless branches, and a time-budgeted lookahead — tested at 5 ms and 25 ms per step, plus a depth-first version — that bounds search by wall-clock, so per-step latency stays predictable for real-time use.
- A clean Prolog-solver / Python-frontend split, migrated off SWI-Prolog’s native XPCE graphics onto the pyswip bridge.