Master’s Thesis: Computer Vision for Perception and Localization in Autonomous Floor Grinding
We are looking for one or two master’s students with an interest in robotics and computer vision to conduct their thesis project with the Floor Grinding team at Husqvarna Construction in Jonsered, Sweden.
Autonomous floor grinders must operate reliably in large and challenging indoor environments. To follow planned grinding paths, achieve consistent surface coverage, and interact safely with their surroundings, the machines require accurate information about their own motion and the surrounding environment.
Existing systems may use sensors such as LiDAR, wheel odometry, and inertial measurement units. However, industrial environments can present difficult conditions, including large open areas, uniform concrete floors, plain walls, changing surroundings, moving equipment, dust, vibration, limited visual texture, and varying illumination. Under these conditions, individual sensors may provide incomplete or uncertain information.
Computer vision could provide additional information that complements the existing sensor system. One possible direction is visual SLAM, or vision-based simultaneous localization and mapping, in which camera data is used to estimate the machine’s motion while simultaneously building or updating a representation of its surroundings. Visual SLAM could be used independently or combined with LiDAR, IMU, and wheel odometry in a multi-sensor fusion solution.
Other possible applications of computer vision include:
•
Visual odometry and motion estimation
•
Detection of moving, temporary, or unsuitable reference objects
•
Recognition of environmental features and landmarks
•
Estimation of sensor alignment and machine parameters
•
Detection of floor boundaries, obstacles, and relevant work-area characteristics
•
Improved sensor fusion and confidence estimation
The central question is how camera-based perception can complement the machine’s existing sensors and improve its ability to localize itself and understand its operating environment.