Copyright ยฉ 2025 Alessio Borgi, Andre Khoo, Kristjan Tarantelli, Rasmus Bรถrjesson Dahlstedt
Certificate of Excellence: 1st Place in the TESP 2025 Competition
Robot Navigation, Obstacle Avoidance and Interaction on the Moon
This project aims to design, build, and program an autonomous mobile robot capable of navigating a static sand terrain, simulating lunar conditions, and interacting with objects in its environment.
A Space Robotics Lab project under Prof. K. Yoshida, Tohoku University
Our task was to design and build an autonomous mobile robot capable of navigating a simulated lunar environment, avoiding obstacles, and interacting with specific target objects placed on the terrain.
This mission required overcoming several key challenges. First, we had to build the robot entirely from scratch, optimizing its mechanical design for a sandy, uneven surface that caused slippage and instability. This meant carefully selecting components, configuring the drivetrain, and iteratively refining the robotโs physical structure.
The second major challenge was to implement a robust navigation and control system. This involved generating reliable paths through the terrainโdespite limited sensor data and unpredictable motionโand finally enabling the robot to detect and interact with objects (turtles) using onboard vision and a custom-built gripper.
The Koopas are stranded on the Moon, in the domain of Dry-Bowser. With Mario on holiday with Peach, it's up to R.O.B. to rescue them by navigating lunar terrain and interacting with targets (Turtles).
|
|
We iteratively prototyped 4 robot models, each improving on mobility, power, and design stability:
- Tsukikage: Lightweight, 2 motors, simple
- Seigetsu: 4 motors, powerful but heavy
- Mikazuki: Compact turning, unstable front
- Tenshiko (Final): Loader-inspired, 1 motor linear gripper
The final robot is the following:
|
|
- Raspberry Pi โ Sensor & ML processing
- EV3 Brick โ Motor control
- Camera Module โ Turtle detection
- Motors โ Controlled via ROS2 or manually
- Input: Binary Map (simulated satellite view)
- Planner: Dijkstra Algorithm (suboptimal/fallback pathing)
- Controller: PD controller outputs velocity
(v)and angular velocity(w) - Visualization: RViz
To enable autonomous navigation, we developed a computer vision and planning pipeline that converts a raw image of the terrain into a structured navigation map used by our path planner.
A top-down image of the terrain is acquired, simulating satellite imagery of a lunar environment with sand and rocks.
The image is converted into a binary map using basic thresholding. Obstacles appear in black, and free space appears in whiteโsimplifying the environment for planning algorithms.
Using the binary map, a distance transform is applied to compute a distance map. This map indicates how far each point is from the nearest obstacle. Brighter regions are safer and more navigable. Finally, we apply a retraction process on the distance map to generate a safe navigation zone, pulling the valid path away from obstacles while preserving reachability. This map serves as the input for our Dijkstra-based path planner.
- Camera: GC0308 CMOS, 2MP, 30FPS
- Dataset: 240 labeled images (manual masks)
- Platform: Roboflow for data augmentation + training
- Model: Simple object detection used for turtle identification and aimpoint guidance
- Issues: Blurry images โ failed detections
- Use visual servoing to keep turtle centered in view.
- Simple heuristic-based steering.
- Loader-style linear gripper (must stay off ground while navigating).
- Difficulty aligning due to nondeterministic slip/glide on sand.
| Name | Affiliation | Country |
|---|---|---|
| Andre Khoo | Nanyang Technological University | ๐ธ๐ฌ Singapore |
| Alessio Borgi | Sapienza University of Rome (AI & Robotics) | ๐ฎ๐น Italy |
| Kristjan Jurij Tarantelli | Sapienza University of Rome (AI & Robotics) | ๐ฎ๐น Italy |
| Rasmus Bรถrjesson Dahlstedt | Chalmers University of Technology | ๐ธ๐ช Sweden |
If you like this world, please, have a look at these other projects I have developed:
- AMR_CleaningRobot: Simulation of a Cleaning Robot with the capability of performing SLAM of the environment the robot is navigating, Planning Trajectories to calculate optimal paths considering static and dynamic obstacles, and Dynamic Obstacle Avoidance to detect and navigate around obstacles.