Project video, 2017
A reconnaissance robot explores and maps a maze it has never seen
A robot is put into a maze nobody has surveyed, with a laser range finder, two cameras and no map. It decides for itself where to go next, and comes out with the floor plan it drew on the way.
The 2017 Peradeniya project video. A holonomic robot with a laser range finder and stereo cameras explores a plywood maze it has never seen, builds the occupancy map in the inset, and finishes with the stitched 3D reconstruction of the arena. The localisation and control segment, which was mine, starts at 0:59.
Key moments
- 0:00Where a robot has to go instead of a person
- 0:16Title, and the department behind it
- 0:24The system: holonomic platform, RGB-D and laser range finder, planner
- 0:32Stereo vision: calibration, disparity, depth, point cloud stitching
- 0:44Exploration planning: smallest gap first, adaptive stepping, neighbours
- 0:58Monte Carlo localisation and the recovered trajectory
- 1:12Building the plywood arena
- 1:24The exploration run, with the occupancy map filling in
- 2:35The stitched 3D map of the arena
- 2:56Supervisors and credits
What you are seeing
The sensing half is stereo, not depth hardware
From 0:32 the video follows the vision pipeline: camera calibration against a checkerboard, disparity maps computed from the stereo pair, depth extracted from those, and the per-view point clouds stitched into one cloud of the arena. A laser range finder supplies the metric ranges the vision is fused with. Sachini Ekanayake built this part.
The planning half decides where to go next
From 0:44 the exploration planning system is shown on the occupancy grid it plans over: the unknown environment as the robot sees it, the smallest gap first rule, adaptive stepping, and the neighbour search that produces the next candidate. Nilanga Thelasingha built this part.
Localisation and control, the part I built
From 0:59 the segment is mine. A particle filter localises the robot in the map it is building, and the video shows the particle cloud collapsing onto a single hypothesis and the recovered trajectory over the arena loop. I was the SLAM and control developer on the project, one of three students.
Then it is put in the maze
The plywood arena is built on camera, and from 1:24 the run is filmed from above with the occupancy map inset in the corner. The map fills in corridor by corridor as the robot commits to one frontier at a time, and the video closes on the stitched 3D reconstruction of the whole arena.
Recording details
- System
- Laser range finder fused with stereo cameras, holonomic platform
- Mapping
- Occupancy grid with particle filter localisation
- Place
- Department of Electrical and Electronic Engineering, University of Peradeniya
- My role
- SLAM and control developer, one of 3 students
- Audio
- Narrated, with the original soundtrack
- Length
- 3:04 (184 seconds)
- Frame
- 1280 x 720 pixels
- File
- MP4, H.264, 11.6 MB
- Published
- September 2026
The work behind it
The project became an ICIIS 2017 paper, and it is the question the PhD asks again ten years later with a learned model and a single camera.
- Autonomous exploration planning strategy for a reconnaissance agent ICIIS 2017
- The same question, ten years later the research arc
Another clip
Watch next
The same job, ten years on, from the air
One camera, one IMU, and a scene graph built over a corridor loop.