Sensor-Based SLAM Navigation — iRobot Create

The Problem

A robot dropped into an unknown room has no map. It can't plan a path to a destination it can't see, and it can't build a map it hasn't explored yet. The only way forward is to do both at once — map the environment while navigating it — using only the readings from its own sensors.

The Challenge

The iRobot Create platform has no built-in mapping capability. The challenge was to mount IR rangefinders on servo motors to achieve 360° coverage, build an occupancy grid map from the noisy sensor readings in real-time, and use that map to plan collision-free paths through a maze — all within the compute and power constraints of an embedded system.

Architecture & System Design

Rotating sensors build real-time 2D map of environment from range data. Occupancy grid algorithm marks free and occupied spaces. Path planning algorithm searches map for collision-free navigation routes. Robot executes waypoints using heading control feedback.

Two servo motors rotated IR rangefinder sensors to sweep the environment and collect range data at discrete angular increments. An occupancy grid algorithm converted range readings to a 2D binary map (occupied/free). The map was iteratively updated as the robot moved, with odometry used for dead-reckoning position estimates. A Rapidly-exploring Random Tree (RRT) planner searched the map for collision-free paths to the goal, with the robot executing waypoints via a proportional heading controller.

Code Walkthrough

3-step walk-through of the production implementation — file paths and intent shown above each block.

1. 01

   Step 1 of 3

   IR range sweep → probabilistic occupancy grid

   slam/occupancy.py

   Each IR rangefinder reading gives range and angle relative to the robot. Bresenham's line algorithm traces the beam through the grid: every cell along the ray gets a small free-space credit, and the terminal cell gets a large occupied hit. Probabilistic updates (add/subtract, clamp 0–1) mean a single noisy reading doesn't immediately flip a cell — the map stabilises over repeated sweeps.

   pythonCopy

   Takeaway

   Probabilistic updates tolerate sensor noise — a single return doesn't hard-set a cell; the map stabilises after 3–4 overlapping sweeps, matching what was observed in practice.

2. 02

   Step 2 of 3

   RRT path planner on the occupancy grid

   slam/rrt.py

   Grid-based search (Dijkstra/A*) is complete but slow on a 200×200 grid with a 3-DOF robot. RRT builds a random tree that biases growth toward unexplored free space, finding a path in milliseconds. A 5% goal-bias nudges the tree toward the target without the greediness that causes local minima in pure goal-directed search.

   pythonCopy

   Takeaway

   The 5% goal bias is the key tuning parameter — too low and the tree explores randomly forever, too high and it gets stuck in local narrow passages between obstacles.

3. 03

   Step 3 of 3

   Waypoint follower — iRobot Create serial execution

   slam/execute_path.py

   The RRT output is a list of (x, y) waypoints in world coordinates. The robot executes them via a proportional heading controller: the bearing error between current heading and the next waypoint drives the turn radius, and odometry from wheel encoder deltas keeps the dead-reckoning position updated without GPS.

   pythonCopy

   Takeaway

   Wrapping bearing error to [-π, π] before the proportional gain prevents the 2π discontinuity from commanding a full reverse spin when the robot overshoots 180°.

Results

The robot successfully mapped lab maze environments and navigated from start to goal positions autonomously. The occupancy grid converged to an accurate representation within 3-4 passes of the environment. RRT path planning consistently found collision-free paths in under 500ms on the embedded hardware.

Gallery & Demos

Occupancy Grid Map — SLAM Output

2D binary map built from 360° IR sensor sweeps showing occupied (walls) and free (navigable) spaces.

Lab Environment 1 — Test Space

First maze-like test environment where the robot performs autonomous mapping and navigation.

Lab Environment 2 — Alternate Configuration

Second test space with different wall layout to validate robustness of the SLAM pipeline.

RRT Path Planning — Collision-Free Route

Rapidly-exploring Random Tree search result showing planned path from start to goal avoiding obstacles.

Planning Map — Cost Grid

Navigability cost grid used by the planner to guide the robot through open space while maintaining clearance from walls.

Click any image or video to expand · ← → keys navigate