Synthesis for Robotics

Transcription

1 Synthesis for Robotics Contributors: Lydia Kavraki, Hadas Kress-Gazit, Stéphane Lafortune, George Pappas, Sanjit A. Seshia, Paulo Tabuada, Moshe Vardi, Ayca Balkan, Jonathan DeCastro, Rüdiger Ehlers, Gangyuan Jing, Morteza Lahijanian, Matthew R. Maly, Matthias Rungger, Kai Weng Wong Speaker: Rüdiger Ehlers ExCAPE Expeditions in Computer Augmented Program Engineering 20 th August 2013 Speaker: Rüdiger Ehlers (UCB/Cornell) Synthesis for Robotics Philadelphia, 20 th August / 11

2 Overview Goal High-level programming of robots by end-users Challenges Physical systems Non-linear dynamics Open worlds Optimality of control Work by Roberto Villalba, photo by R. Ehlers Speaker: Rüdiger Ehlers (UCB/Cornell) Synthesis for Robotics Philadelphia, 20 th August / 11

3 Overview Goal High-level programming of robots by end-users Challenges Physical systems Non-linear dynamics Open worlds Optimality of control Work by Roberto Villalba, photo by R. Ehlers Speaker: Rüdiger Ehlers (UCB/Cornell) Synthesis for Robotics Philadelphia, 20 th August / 11

4 A Challenge Problem z 3 z 5 z 4 Table z 1 Kitchen z pickup z 0 Robot Waiter This challenge problem is: simple to understand, yet challenging, scalable, reactive, and z 2 has mission planning and hybrid control aspects. Speaker: Rüdiger Ehlers (UCB/Cornell) Synthesis for Robotics Philadelphia, 20 th August / 11

5 Challenge problem video Speaker: Rüdiger Ehlers (UCB/Cornell) Synthesis for Robotics Philadelphia, 20 th August / 11

6 Pushing the Frontier in Synthesis for Robotics 1 What to do if input/output spaces are infinite? Synthesis with Identifiers (UC Berkeley, Cornell) 2 What if our sensors and actuators are imprecise? Robust Synthesis (UCLA, Cornell) 3 What if the robot s task is not 100% achievable? Iterative Motion Planning for Hybrid Systems (Rice, Cornell) 4 What if we are interested in efficient solutions? Cost-optimal Synthesis (Cornell) 5 What if the environment behaves faulty? Error-resilient Synthesis (Cornell) Speaker: Rüdiger Ehlers (UCB/Cornell) Synthesis for Robotics Philadelphia, 20 th August / 11

7 Synthesis with Identifiers.##.# true order/a := itemordered start q 0 q 1 deliver q 2 deliver q 3 Main idea deliver itemdelivered a true Allow checking the realizability of many specifications that cannot be implemented in a finite-state fashion Synthesize a memory-conservative implementation in case of realizability ## # # Speaker: Rüdiger Ehlers (UCB/Cornell) Synthesis for Robotics Philadelphia, 20 th August / 11

8 Robust Synthesis High-level controller Motion layer true/m 1 q 0 q 1 i/m 2 true/ i/m 3 m 2 q 2 Challenges Uncertainty in measurement Imprecision in motion y+w ϕ+w x+ W Figure by Matthias Rungger Speaker: Rüdiger Ehlers (UCB/Cornell) Synthesis for Robotics Philadelphia, 20 th August / 11

9 Iterative Motion Planning for Hybrid Systems Temporal Logic Specification High Level Planner abstraction Synergy Layer Hybrid System with Dynamics Sampling-based Motion Planner Braking and Re-abstraction obstacle discovered Continuous Solution Trajectory Reference: Matthew R. Maly, Morteza Lahijanian, Lydia E. Kavraki, Hadas Kress-Gazit, Moshe Y. Vardi: Iterative temporal motion planning for hybrid systems in partially unknown environments. HSCC 2013: Speaker: Rüdiger Ehlers (UCB/Cornell) Synthesis for Robotics Philadelphia, 20 th August / 11

10 Cost-optimal Synthesis r 1 r 2 r 3 r 4 r 5 r 6 r 7 r 8 r 12 r 9 r 10 r 11 Novelty: Use a two-dimensional cost notion to track waiting cost and transition cost in adversarial environments. Reference: Gangyuan Jing, Rüdiger Ehlers, and Hadas Kress-Gazit: Shortcut Through an Evil Door: Optimality of Correct-by-Construction Controllers in Adversarial Environments. IROS 2013 Speaker: Rüdiger Ehlers (UCB/Cornell) Synthesis for Robotics Philadelphia, 20 th August / 11

11 Error-resilient Synthesis Aim Let the robot behave reasonably in cases of environment assumption violations LTop Classroom RTop Solution Door1 Obstacle Door2 Modify the strategy extraction part of the synthesis algorithm let it tolerate moving away from the goal whenever there is no alternative. LBottom Playground RBottom Speaker: Rüdiger Ehlers (UCB/Cornell) Synthesis for Robotics Philadelphia, 20 th August / 11

12 Error-resilient Synthesis Classroom Classroom Classroom Classroom LTop RTop LTop RTop LTop RTop LTop RTop Door1 Obstacle Door2 Door1 Obstacle Door2 Door1 Obstacle Door2 Door1 Obstacle Door2 LBottom RBottom Playground LBottom RBottom Playground LBottom RBottom Playground LBottom RBottom Playground Door1 closed Door2 stays Door1/Door2 open status is ignored Speaker: Rüdiger Ehlers (UCB/Cornell) Synthesis for Robotics Philadelphia, 20 th August / 11

13 Summary General goal Enable synthesis of reasonable robot controllers for reactive scenarios with correct-by-construction hybrid control by end users, e.g. a robot waiter Approaches to push the frontier presented in this talk 1 Synthesis with Identifiers 2 Robust Synthesis 3 Iterative Motion Planning for Hybrid Systems 4 Cost-optimal Synthesis 5 Error-resilient Synthesis Speaker: Rüdiger Ehlers (UCB/Cornell) Synthesis for Robotics Philadelphia, 20 th August / 11