Abstract:
"Human Following Robots (HFR)s have gained significant attention in recent years due to their potential applications in various domains, such as personal assistance, security, and entertainment.
However, one of the significant challenges HFRs face is handling occlusions, where the
human subject is partially or completely obstructed from the robot’s sensors. This research
project addresses this challenge by developing an occlusion-aware perception system for HFRs.
The proposed system integrates data from wearable trackers and indoor localization systemsto accurately map the environment and track the human subject’s movements. By combining multiple input sources, the system aims to overcome the limitations of vision-based tracking methods and effectively handle scenarios of complete occlusion. The research project involves designing algorithms for handling complete occlusion, implementing a simulated custom wearable tracking device, and building simulated environments for testing and validation.
Testing results reveals that the runtime for occlusion recovery increases with the number of static obstacles. In unknown environments, the average recovery time ranges from 10.06 seconds for 2 obstacles to 24.91 seconds for 9 obstacles, while in known environments, it improves to 7.09 seconds for 2 obstacles and 17.08 seconds for 9 obstacles. Accuracy testing demonstrates that the system maintains high success rates, with 100% successful recoveries and average times between 6.08 and 6.64 seconds for Ultra-Wideband (UWB) error margins of 5% to 10%. However, the recovery success rate drops to 80% when the error margin increases to 12% and 15%. These results underscore the system’s potential in enhancing occlusion handling while highlighting areas for further refinement in diverse scenarios."
