Charles Tang

EECS Department, University of California, Berkeley

Technical Report No. UCB/EECS-2021-61

May 13, 2021

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2021/EECS-2021-61.pdf

Modern day robots deployed in the ground or sky have to frequently navigate a priori unknown environments. In these scenarios, robots need to make goal-driven decisions while also satisfying safety constraints imposed by obstacles. Guaranteeing safe operation in unknown environments with limited-range sensors remains a challenging problem for autonomous vehicles. Recent work proposed a Hamilton-Jacobi reachability framework for computing safe operational regions for such scenarios. Unfortunately, controllers synthesized from this framework are jerky and uncomfortable when deployed on the robot. Furthermore, evaluating HJI algorithms in real-time scenarios remains difficult due to the curse of dimensionality. In this work, we implement the BEACLS ROS toolbox which allows for real time computation of HJI safety sets for robotic navigation tasks. The state-of-the-art algorithms are benchmarked against MATLAB implementations and demonstrate a 6x speedup in computation time. We then explore bang bang control, unconstrained optimization, and constrained optimization formulations to blend the safety set and robot motion planner. We conclude by evaluating each blending scheme’s ability to produce smooth, safe, feasible, and goal reaching trajectories for known and unknown environment point-goal navigation tasks.

Advisors: Claire Tomlin

---

BibTeX citation:

@mastersthesis{Tang:EECS-2021-61,
    Author= {Tang, Charles},
    Title= {Real-time Robotic Safety Set Blending Schemes},
    School= {EECS Department, University of California, Berkeley},
    Year= {2021},
    Month= {May},
    Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2021/EECS-2021-61.html},
    Number= {UCB/EECS-2021-61},
    Abstract= {Modern day robots deployed in the ground or sky have to frequently navigate a priori unknown environments. In these scenarios, robots need to make goal-driven decisions while also satisfying safety constraints imposed by obstacles. Guaranteeing safe operation in unknown environments with limited-range sensors remains a challenging problem for autonomous vehicles. Recent work proposed a Hamilton-Jacobi reachability framework for computing safe operational regions for such scenarios. Unfortunately, controllers synthesized from this framework are jerky and uncomfortable when deployed on the robot. Furthermore, evaluating HJI algorithms in real-time scenarios remains difficult due to the curse of dimensionality. In this work, we implement the BEACLS ROS toolbox which allows for real time computation of HJI safety sets for robotic navigation tasks. The state-of-the-art algorithms are benchmarked against MATLAB implementations and demonstrate a 6x speedup in computation time. We then explore bang bang control, unconstrained optimization, and constrained optimization formulations to blend the safety set and robot motion planner. We conclude by evaluating each blending scheme’s ability to produce smooth, safe, feasible, and goal reaching trajectories for known and unknown environment point-goal navigation tasks.},
}

EndNote citation:

%0 Thesis
%A Tang, Charles 
%T Real-time Robotic Safety Set Blending Schemes
%I EECS Department, University of California, Berkeley
%D 2021
%8 May 13
%@ UCB/EECS-2021-61
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2021/EECS-2021-61.html
%F Tang:EECS-2021-61