A multi-robot path planning demo built with Q-Learning in MATLAB. Multiple robots navigate an N × N grid to reach a common goal while avoiding static obstacles. Grid size, number of robots and obstacles are user-specified at runtime. The project demonstrates reinforcement learning basics, collision handling, and a simple visualization / animation of robot trajectories.
This project implements a baseline multi-robot planner where each robot learns a local policy using Q-learning on the discrete grid. Robots receive positive reward on reaching the goal and negative reward for hitting obstacles. After training, the learned Q-table is used to generate robot trajectories which are animated on a 2D grid.
- Language: MATLAB (script)
- Paradigm: Reinforcement learning (Q-Learning) on discrete grid
- Input: Grid size, number of robots, number of obstacles (entered by user)
- Output: Animated simulation, summary metrics (avg path length, computation time, collisions, Q-table memory size)
Random placement of robots, goal, and obstacles. Q-Learning training loop with tunable hyperparameters:
-
gamma (discount factor)
-
alpha (learning rate)
-
epsilon (exploration probability)
-
episodes (training episodes)
-
Action set: Right, Left, Up, Down (4-neighbour moves).
-
Visualization of grid, obstacles, goal, robot positions and animated trajectories.
-
Basic metrics printed at the end: average path length, computation time, number of collisions, model size.
-
Initialize grid and randomly place goal, obstacles and robots.
-
Initialize Q-table: Q_table(rows, cols, 4) (one value per action per cell).
-
Train via episodes:
For each robot per episode, choose action by ε-greedy.
Update Q-values using the Q-learning update:
Q(s,a) ← (1 − α)Q(s,a) + α (reward + γ max_a' Q(s',a'))
-
Rewards: large positive for reaching goal, negative for obstacle, small negative per step.
-
After training, generate trajectories by greedily following the learned policy (argmax Q).
-
Animate robot movement and report metrics.
Enter Grid Size as [rows cols]: [10 10]
Enter Number of Robots: 5
Enter Number of Obstacles: 10
Generating grid...
Plotting...
Path Length: 7.00
Computation Time: 0.2573 seconds
Collisions: 0
Max Robots: 5
Model Size: 3200 bytes
The following is the visualisation for the given input.
The blue lines higlight the trackers for each robot which has started from random positions, taking in the most appropriate path based on Q-Learning, to reach their common goal at (9,2). The square blocks are the obstacles.