IoT FPGA System : Maze Runner & Pac-Man

A two-game IoT platform built on a pair of PYNQ-Z1 FPGA boards connected via an AWS EC2 relay server, with all live game state projected onto the floor through a browser-based interface.

Game 1 : Maze Runner: a 10×10 maze is generated in hardware using Kruskal's algorithm, solved by a BFS (or A*) hardware accelerator running entirely in programmable logic, and a physical rover navigates the shortest path. After the first run, three blue objects placed on wall cells are detected by the camera node and the maze is dynamically updated, triggering a re-solve and a new (potentially shorter) path.

Game 2 : Pac-Man: the rover becomes a user-controlled agent navigating a 7×13 arena via a wireless controller, collecting six coins while evading up to four ghost agents whose pursuit logic : line-of-sight detection and wall-following and runs entirely in FPGA fabric, with ghost positions relayed to the server at sub-100 ms round-trip latency.

---

Repository Structure

Folder	Description
bfs_search/	BFS hardware accelerator: SystemVerilog RTL, pipelined distmap + backtrace + FSM, AXI-Stream wrapper, Verilator testbench, Vivado bitstream.
aStar search/	A* hardware accelerator: SystemVerilog RTL, 4-state FSM, priority queue, neighbour expansion, backtrace, AXI-Stream wrapper, multi-maze testbench, Vivado bitstream.
ghost_new_ip/	Ghost AI hardware: single ghost_top instance (LOS + wall-follow 3-stage pipeline), 4× parallel ghost_multi_axi_lite wrapper, testbench, pre-built bitstream and HWH file.
mazegen_hw/	Kruskal maze generation HLS IP: generates a unique 10×10 perfect maze each run via union-find. Includes bitstream and Jupyter notebook.
camera/	Camera pipeline notebook (Video-Final.ipynb): AXI VDMA + hardware downsampler, red-dot homography calibration, blue-blob detection, grid coordinate POST to AWS.
rover/	Rover control notebooks: Game 1 full pipeline (GAME1-Copy1.ipynb) and Game 2 Pac-Man controller (Pacman_movement_controller.ipynb). Includes MicroBlaze I2C IMU code and stepper motor control.
Hamming_code/	Hamming(7,4) integrity layer (hamming_maze.py): encodes maze rows as 7-bit codewords, detects and corrects 1-bit errors before BFS runs.
extension_parallel_solving/	Extension: parallel maze solving experiment. See folder README for details.
voice/	Voice control unit: continuous 3-second audio recording loop, speech recognition, and server-side page navigation via /set_page.
website code/	AWS EC2 Flask server (mission_control.py) and all web pages (index.html, fpga_demo.html, fpga_arena.html, archives.html).

---

Getting Started

1. AWS Server

cd "website code"
pip install flask flask-cors boto3
python mission_control.py

The server runs on port 5000. Ensure AWS credentials are configured with access to the six DynamoDB tables: MazeTable, RoverLogs, GhostLogs, VisionPositions, Leaderboard, VoiceCommands.

2. Node B — Rover (Game 1)

Open rover/GAME1-Copy1.ipynb on the rover PYNQ board and run all cells in order:

1. WiFi connection
2. Kruskal maze generation (mazegen_hw/)
3. Maze POST to AWS → projection updates
4. BFS path computation (bfs_search/)
5. Physical rover navigation
6. Blue-dot wall removal and re-solve

3. Node A — Camera & Ghost (Game 1 / Game 2)

• Game 1: open camera/Video-Final.ipynb and run. The notebook calibrates the perspective transform using four red corner markers, then continuously detects blue objects and POSTs grid coordinates to AWS.
• Game 2: open ghost_new_ip/ghosts_newmaze.ipynb and run. Loads ghost_bd.bit, syncs the arena to AWS, and starts the ghost control loop.

4. Node B — Rover (Game 2)

Open rover/Pacman_movement_controller.ipynb. Use the wireless controller input loop to move the rover; each move is validated server-side before physical execution.

---

Hardware Testing (Simulation)

Each hardware module has its own README with simulation instructions. For example:

A*:

cd "aStar search"
iverilog -g2012 -o sim_astar astar_top.sv astar_fsm.sv astar_init.sv update_block.sv backtrace.sv astar_multi_tb.sv && vvp sim_astar