Improving Human Safety through Integrated Autonomous Motion Planning
Loading...
Author(s)
Iyengar, Divya
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
Workers in dangerous, unstructured environments face an increased risk of injury from collisions with heavy equipment or moving objects. To improve human safety in these scenarios, visual feedback is recommended to enhance a person’s situational awareness and reduce the number of collisions. Previous work on human-centric approaches have used either global planning around static obstacles or local dynamic obstacle dodging to address this research question. In our proposed solution, we combine a global RRT* planner, leveraging a Dubins path tuned for human performance, with a local HCVO planner. Our human-based, integrated planning algorithm directly maps human movement to provide safe navigation suggestions to the user. First, we introduce the components of the integrated planner, tune algorithm parameters, and conduct a pilot experiment to develop a human dynamics model. Then, we build a simulated, complex environment with different classes of static obstacles and design a dynamic obstacle AI strategy. Finally, we conduct human subject studies on N = 10 subjects in virtual reality to test the proposed algorithm in the designed environment across 48 trials. The algorithm’s performance is compared to a baseline condition, which includes an overhead minimap camera that emphasizes threats in the environment across 20 levels containing randomized obstacle locations. Our results quantify the number of collisions and reaction time as the primary performance metrics to evaluate the effect of the integrated planning algorithm on human safety. Significant benefit is observed in the number of static obstacle collisions (p = 0.01) and reaction time (p = 0.004). For total collisions, the algorithm performs better than the baseline condition on average. Additionally, the percent of dynamic collisions over the total possible dynamic obstacle hits is on average better in the algorithm condition against the baseline condition. This work demonstrates that 1) human-based motion planning can improve human navigation around static obstacles in complex environments and 2) online path suggestions augment human’s situational awareness for dynamic obstacle avoidance.
Sponsor
Date
2023-12-13
Extent
Resource Type
Text
Resource Subtype
Thesis