Featured Robotics Project

Escape Challenge

Autonomous exploration + mapping + escape behavior using obstacle distance grids, A* planning, frontier exploration, and a simple state machine.

ROS2 Occupancy Grid Obstacle Distance Grid A* Frontiers State Machine
Escape Challenge hero

What it does

Explores a maze, builds a map, returns home, then escapes after an “edge removal”.

Core algorithms

Obstacle distance grid + 8-way A* + frontier detection + centroid navigation.

Safety

Blocks cells near obstacles and biases paths away from walls.

Obstacle Distance Grid

Turns raw mapping into “safe-to-traverse” planning space.

Distance Grid / Map
Obstacle distance grid visualization
Map representation used for safe navigation (obstacles expanded to account for robot size).

Key details

  • Assumes the environment is mostly static while mapping.
  • Models the robot as a point by “growing” obstacles to handle robot footprint.
  • Uses 8-way expansion so diagonal travel is allowed (efficient paths).

A* Planning

Fast, direct paths using an 8-connected grid.

Heuristic

Euclidean distance pairs naturally with 8-way expansion and supports diagonal shortest paths.

  • Considered alternatives: Manhattan (4-way) and diagonal distance heuristics.
  • Rotation cost could be added by weighting turn magnitude into g(x).

Reality vs tests

Some scripted cases may fail even if real-world runs succeed due to sensor noise + replanning.

Frontier Detection

Frontiers are the boundary between known free space and unknown space.

Frontier Visualization
Frontier visualization
Example: frontier cells clustered into connected frontiers (choose one to explore).

Definition + growth

  • A cell is a frontier if it’s free and adjacent to unknown space.
  • Frontier cells are grown/clustered into connected components.
  • Navigates toward the middle to maximize new mapped area.

Limitation

Noise can create tiny “fake” frontiers. A gain-based method could prioritize higher information gain targets.

Exploration Behavior

Why centroid navigation works—and where it breaks.

  • Centroids may be unreachable or sit inside tight geometry.
  • Doesn’t prioritize closest reachable frontier points (can waste time/energy).
  • Treats frontiers “equally” instead of choosing the most promising region.

Safety + Parameters

Hard constraints + soft bias away from obstacles.

Hard safety block

Cells within 0.05 m of occupied cells are treated as blocked.

Soft safety bias

A safety weighting factor of 0.6 biases the path away from obstacles.

Escape Logic

After edge removal, re-explore and drive toward the farthest frontier to exit.

State Machine
Escape state machine diagram
Initializing → Exploring → Returning Home → Escape.

Behavior

  • Reset map after detecting edge removal (challenge-allowed approach).
  • Navigate toward the largest / farthest frontier first to escape.
  • Kept the Lab 12 state machine for clarity and reliability.

Reflection

Biggest improvements: better tuning/testing of safety distance (to prevent wall clips), and more principled detection of wall removal (phase out old map values instead of full reset).