Maze Navigation
Maze navigation using a unicycle model with forward speed and yaw-rate controls. The examples compare density feedback, density filters, and density MPC while following A* waypoints through cluttered layouts with rectangular wall-density obstacles.
Run
Run one maze controller; change the scenario or controller script to compare results.
python examples/maze/density_feedback.py --scenario wide
python examples/maze/density_filter.py --scenario rooms
python examples/maze/density_mpc.py --scenario roomsSystem Dynamics
The maze examples use the same unicycle state and controls as the non-holonomic page. A sparse A* path supplies waypoint references, while each occupied maze cell is converted into a rectangular wall obstacle inside the density function.
Pick a maze and density based controller. The orange markers show the waypoint reference, the green curve shows the nominal baseline trajectory, and the blue curve shows the density controller output.
Maze 2 - Density MPC
- Status
- success
- Steps
- 142
- Min clearance
- 0.022
Density Filter vs Density MPC
Maze 4 is the one default case where the one-step density filter collides while density MPC reaches the goal. The filter is greedy: it enforces a next-step density condition while staying close to the nominal command. In the dense room layout, that local correction is not enough to plan the turn sequence. MPC is predictive: it evaluates a short horizon of future states, so it can trade immediate tracking against future safety and steer through the room.
Maze 4 - Density filter
One-step filter: local correction collides in the dense room layout.
Maze 4 - Density MPC
Predictive MPC: succeeds by reasoning over a short future horizon.
All Maze Configurations
Each row shows density feedback, density filter, and density MPC for a fixed maze. All examples use the same sparse, clearance-aware waypoint profile; Maze 4 highlights the filter failure case.