Abstract
A robot moving through an unfamiliar building must build a picture of its surroundings and work out where it sits within them at the same time. Solving both problems together from camera images is known as SLAM, short for simultaneous localization and mapping. The cameras used here record a normal colour image together with the distance to everything in view; such devices are called depth cameras, or RGB-D cameras, and they suit indoor robots because they provide both appearance and scale without the cost of a laser scanner. Many open-source SLAM systems exist, and they differ considerably in how they store the map, how they recognize a place already visited, how much computing power they require, and how well they cope with dim light or rapid motion.
Supervision
- Prakash Aryan
- Sebastiano Panichella
Motivation
Published results are usually reported on standard benchmark datasets, and those rankings do not always carry over to a real low-cost robot. Camera calibration, poorly synchronized image streams, vibration from the wheels, blur during turns, blank walls with little texture, and a small onboard computer can all change which system performs best. A controlled comparison on the same physical robot, in the same building, therefore provides practical evidence that published numbers cannot.
Goal
The student will set up and compare two widely used open-source SLAM systems, RTAB-Map and ORB-SLAM3, on a small wheeled research robot called the TurtleBot3, fitted with a depth camera from the Intel RealSense family. Both systems will be given exactly the same recorded sensor data and evaluated on identical indoor routes. The objective is not to develop a new SLAM algorithm, but to produce a reproducible comparison and clear guidance on which system suits which conditions.
Project Scope
- Build one software pipeline, using the standard robotics framework ROS 2, that records colour images, depth, camera calibration, wheel odometry, and robot positions in step with each other, saving them as reusable recordings so both systems receive identical input.
- Configure both SLAM systems and inspect their maps and estimated paths in the standard robotics visualization tool, RViz, using the same map resolution and comparable settings wherever possible.
- Drive an office loop and a corridor loop under normal lighting, then repeat one route under a harder condition such as dimmed lights or untextured walls, running each experiment three times.
- Provide launch files, evaluation scripts, and concise setup notes so that the comparison can be repeated on another robot of this kind.
Evaluation
- Accuracy: how far the estimated path drifts from the true one, and whether the robot ends where it started after completing a loop.
- Map quality: how well walls and major objects align with a measured floor plan, together with map completeness and visible distortion.
- Performance: processor and memory use, processed frame rate, dropped frames, and the time required to start up.
- Robustness: how often each system loses tracking, how often it recovers, and how many runs complete.
Requirements
The student should be comfortable with Python or C++, Linux, and the basics of ROS 2. Previous SLAM experience is useful but not expected, as most of the work involves careful system integration, repeatable experiments, and honest analysis rather than training models.
Pointers
- C. Campos et al., “ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual-Inertial, and Multimap SLAM,” IEEE Trans. Robotics, vol. 37, no. 6, 2021. One of the two systems compared here.
- M. Labbé and F. Michaud, “RTAB-Map as an open-source lidar and visual SLAM library,” J. Field Robotics, vol. 36, no. 2, 2019. The other system.
- A. Merzlyakov and S. Macenski, “A Comparison of Modern General-Purpose Visual SLAM Approaches,” in Proc. IEEE/RSJ IROS, 2021. A comparison study close in spirit to this project.
- J. Sturm et al., “A Benchmark for the Evaluation of RGB-D SLAM Systems,” in Proc. IEEE/RSJ IROS, 2012.
- Z. Zhang and D. Scaramuzza, “A Tutorial on Quantitative Trajectory Evaluation for Visual(-Inertial) Odometry,” in Proc. IEEE/RSJ IROS, 2018. Describes how path error should be measured.