Execution and analysis of high-level tasks with dynamic obstacle anticipation

Experimental Evaluation and Formal Analysis of High‐Level Tasks with Dynamic Obstacle Anticipation on a Full‐Sized Autonomous Vehicle

“…This paper continues previous work (Johnson et al., ), and investigates, via experimental data collected from a full‐scale autonomous vehicle, the fusion of probabilistic perception (through the anticipation of dynamic obstacles) with deterministic decision making (by a correct‐by‐construction hybrid controller). The anticipation algorithms provide a probabilistic belief about the future state of the environment, allowing the robot to plan ahead rather than relying entirely on reactionary planning.…”

“…In this paper, as in Johnson and Kress‐Gazit () and Johnson et al. (), the probability that the robot satisfies its task is found with respect to those errors in the robot's observations.…”

“…The work presented in this paper directly addresses the problem of incorporating probabilistic environmental information into a deterministic, synthesized controller. Unlike the authors' previous work (Johnson et al., ), in this paper the anticipation algorithm and synthesized controller are implemented on a full‐scale autonomous car, pictured in Figure , which operated in a controlled environment with multiple simulated vehicles, which were controlled by human drivers (also shown in Figure ). Furthermore, experimental data were collected using the autonomous vehicle and human‐controlled obstacles (rather than via simulation), and the results are compared with a formal analysis of the synthesized controller, showing good agreement between the collected data and the output of the probabilistic model.…”

“…Anticipation of dynamic objects has received interest in the field in recent years, and fully probabilistic, fast, and accurate algorithms are available (Choi et al., ; Du Toit & Burdick, ; Fulgenzi, Spalanzani, & Laugier, ; Hardy & Campbell, ; Havlak & Campbell, ; Ohki, Nagatani, & Yoshida, ; Ziebart et al., ). Integration of anticipation and planning can be done at a low level, by including anticipation as a term in a cost function while computing vehicle actuation commands, as in Hardy and Campbell (), or at a high level, using anticipation to choose a route, as in previous work by Johnson, Havlak, Campbell, and Kress‐Gazit ().…”

Hot-swapping robot task goals in reactive formal synthesis

“…As a precursor to recent work concerning relaxations of these restrictions, the degradation of such "perfect-world" controllers in the presence of sensing uncertainty is explored in [10]. Examples of relaxations considered in recent work include weakened time synchronization requirements in distributed applications [4], online changes to the workspace cell decomposition [16], and mapping of initially unknown planar workspaces (assuming perfect localization) [20], [23].…”