Collision Avoidance in Human-Robot Collaborative Workspaces using multiple low-cost 3D Cameras
- Collision Avoidance in Human-Robot Collaborative Workspaces using multiple low-cost 3D Cameras
- Salzburg Research
- Main Task
- With the advent of inexpensive 3D sensors and open source robotics software, low-cost HRC-systems are becoming more competitive against expensive and proprietary systems. For such systems to be safe, the cobot must sense obstacles before touching them. Strict technical requirements exist to realize a safe and efficient system to work with, such as low latency, high accuracy and low memory consumption. This bachelor thesis proposes a low-cost and open source prototype to enable collision detection for an industrial cobot inside a HRC-workspace. The 3D images of two ceiling-mounted cameras are converted into a occupancy voxel grid and the latency together with the memory consumption is analyzed during conversion. Besides, different comparisons between real-world objects and their voxel grid counterpart are conducted to find out the accuracy. Since the voxel grid includes voxels from the cobot itself, the cobot can be misinterpreted as an obstacle and can thus be inoperable. Those voxels are removed by a filter, whose parameters are first optimized before measuring its efficiency. At last, the voxel grid together with a cobot 3D model are visualized inside a virtual workspace.
- A low latency is reached, and the memory consumption decreases sharply in the tests after optimizing the OctoMap parameters. Altough the measured accuracy is not suited for fine-grain tasks like the gripping of objects with the cobot itself, it is good enough for the detection of obstacles, which is the main goal of this work. Considering those results, together with the filter removing all cobot voxels, the prototype is suited for collision aware path planning. All in all, the goals of this thesis are reached, and the prototype represents a solid foundation for a HRC-system in which humans can safely work together with cobots in close proximity.