Autonomous Mobile Robots and Intelligent Control Issues. Sven Seeland

Transcription

1 Autonomous Mobile Robots and Intelligent Control Issues Sven Seeland

2 Overview Introduction Motivation History of Autonomous Cars DARPA Grand Challenge History and Rules Controlling Autonomous Cars MIT Talos Overview Intelligent Control Systems Controlling Talos 2

3 Mobile Robots Motivations Can work under hostile environmental conditions Can move in confined spaces Expendable in dangerous situations 3

4 Autonomy Definition 1 Autonomy refers to systems capable of operating in the real-world environment without any form of external control for extended periods of time. George A. Bekey, Autonomous Robots: from biological inspiration to implementation and control 4

5 Autonomy Definition 2 A fully autonomous robot has the ability to Gain information about the environment. Work for an extended period without human intervention. Move either all or part of itself throughout its operating environment without human assistance. Avoid situations that are harmful to people, property, or itself unless those are part of its design specifications. - Wikipedia, Autonomous robot 5

6 Autonomy Motivations Remote control might be infeasible Area too large or cluttered for wired control Poor wireless reception No operator required Cheap operation of many units No set working hours No fatigue 6

7 Autonomous Mobile Robots Applications Clearing an area of landmines, bombs and other explosives Rescue robots Service Robots Maintenance Robots Exploration Toys Automated Driving... 7

8 Autonomous Cars Motivation Public Transport Safer Driving More comfortable traveling Delivery Tasks... 8

9 Overview Introduction Motivation History of Autonomous Cars DARPA Grand Challenge History and Rules Controlling Autonomous Cars MIT Talos Overview Intelligent Control Systems Controlling Talos 9

10 Autonomous Cars History PROMETHEUS Project ( ) Initiated by the European Commission PROgraMme for a European Traffic of Highest Efficiency and Unprecedented Safety $1 billion funding Most prominent results where VaMP and ARGO 10

11 Autonomous Cars History VaMP (1995) Versuchsfahrzeug für autonome Mobilität PKW >2000 km from Munich to Kopenhagen and back in normal traffic Up to 180 km/h Up to 158 km without human intervention Mean distance between human interventions: 9 km Lane changes Vehicle passing Active computer vision Radar 11

12 Autonomous Cars History ARGO Project (1998) 2000 km Tour through Italy Above 90% of the time in automatic mode Longest distance without intervention: 54.3 km Two cameras 200 MHz Pentium MMX 12

13 Overview Introduction Motivation History of Autonomous Cars DARPA Grand Challenge History and Rules Controlling Autonomous Cars MIT Talos Overview Intelligent Control Systems Controlling Talos 13

14 DARPA Grand Challenge - History Motivation: Make one-third of ground military forces autonomous by 2015 Off-road tracks 2004: 241 km $1 Million prize money no winner Best vehicle travelled 11,78 km 2005: 213 km $2 Million prize money 5 vehicles succeed All but one got past the maximum distance of

15 DARPA Urban Challenge 2007 Rules 1 $2 Million, $1 Million and $ prizes Complete 60 miles in 6 hours to finish the race Urban environment Decommissioned Air Force Base Street network in residential area Several dirt roads Obey traffic laws All cars on the course at the same time 3 individual missions per car No pedestrians or other moving objects Time penalties for dangerous or erroneous behavior 15

16 DARPA Urban Challenge 2007 Rules 2 One Route Network Definition File (RNDF) Handed out 24 hours before the race Similar to maps used in GPS navigation systems Contains road positions, number of lanes, intersections, parking space locations in GPS coordinates One Mission Description File per Team and Mission (MDF) Handed out on the day of the event Contains a list of checkpoints from the RNDF that the vehicle needs to cross 16

17 Overview Introduction Motivation History of Autonomous Cars DARPA Grand Challenge History Rules Controlling Autonomous Cars MIT Talos Overview Intelligent Control Systems Controlling Talos 17

18 MIT Talos 18

19 MIT Talos Design Considerations Many low-cost sensors Increases perception robustness More complete coverage Higher efficiency in a multi-processor environment Minimal reliance on GPS data Highly distributed computer Better reaction times Downside: higher power consumption Simple low level controls Improve robustness Minimal sensor fusion / asynchronous sensor update 19

20 MIT Talos Specifications Land Rover LR3 Human drivable EMC AEVIT drive-by-wire system 6000 Watts power generator 2 ruggedized UPS Blade Cluster (10 x 4 64-bit CPUs, 2.3 GHz each) Velodyne HDL-64 LIDAR (3D) 12 SICK LIDARs (2D) 5 cameras (752x480, 22.8 images per second) 15 millimeter wave radars Applanix navigation solution (GPS, inertial measurement unit and wheel encoder) 20

21 Overview Introduction Motivation History of Autonomous Cars DARPA Grand Challenge History Rules Controlling Autonomous Cars MIT Talos Overview Intelligent Control Systems Controlling Talos 21

22 Intelligent Control Systems Tasks Actuation Collision avoidance Path-finding / Trajectory planning Mission planning Localization 22

23 Intelligent Control Systems Challenges Uncertainty Dynamic environment Perception Actuation Efficiency Short reaction times in a dynamic environment Limited processing power due to limited space on the moving platform Scalability Potentially huge environment Potentially long operating times 23

24 Intelligent Control Systems Requirements Robustness Input is likely to be inaccurate, incomplete or wrong Unforeseen conditions are likely to occur Speed Quickly react to situations Initial assumptions may be invalid by the time the deliberation process is finished Versatility Multitude of tasks need to be executed simultaneously Highly diverse nature of tasks 24

25 Intelligent Control Systems Basics Control Systems consist of: Input Controller Output Input Controller Output 25

26 Intelligent Control Systems Reactive Systems Purely reactive systems: No planning or learning No internal state Complexity of tasks is limited Highly robust Very quick reaction times Perception Decision Action World 26

27 Intelligent Control Systems Deliberative Systems Deliberative Systems Allows for planning and learning Internal world model Can perform very complex tasks Not very robust Slow Programming Model Building Knowledge Decision Making Reasoning Perception Decisions Actions World 27

28 Intelligent Control Systems Hybrid Systems 1 Hybrid Systems Combination of systems Reactive systems for short term reactions and low level controls Deliberative systems for planning and coordination Model Building Programming Knowledge Decision Making Reasoning Perception Decisions Actions World 28

29 Intelligent Control Systems Hybrid Systems Oftentimes organized in three layers Planning layer handles long time action plans Sequencing divides long term goals into smaller steps Controlling translates those steps into actual actuator commands Planning Sequencing Layers operate in parallel and independently Low layers can fail and report failure to higher layers Controlling Higher layers tend to use deliberative approaches Lower layers tend to use reactive approaches 29

30 Overview Introduction Motivation History of Autonomous Cars DARPA Grand Challenge History Rules Controlling Autonomous Cars MIT Talos Overview Intelligent Control Systems Controlling Talos 30

31 MIT Talos Control System Architecture 31

32 MIT Talos IPC Infrastructure LCM lightweight communications and marshaling Minimalist system for real-time applications Developed specifically for Talos Based on UDP-Multicast Publish/subscribe message-passing model Logging made extremely easy Freely available 32

33 MIT Talos Control System Architecture 33

34 MIT Talos Navigator 1 Highest level of abstraction General planning component Route planning Intersection handling (precedence, crossing, merging) Passing Blockage replanning Turn signaling Failsafe timers Inputs: MDF, lane information, vehicle pose Outputs: goals for motion planner Short term goals within 40-50m range Goal is moved according to the high level intentions Timing of the goal-setting used to control motion 34

35 MIT Talos Navigator 2 Reevaluates situation at 2 Hz Dynamic replanning comes for free Passing Goal remains unchanged Checks if other lane exists and is free Allows the motion planner to use the other lane Two timers for global problem solving: Failsafe timer Progressively sets and unsets global failsafe states Failsafe states progressively relax security constraints Blockage time Determines traffic jams and roadblocks Only works in two-lane roads where a u-turn is possible 35

36 MIT Talos Control System Architecture 36

37 MIT Talos Drivability Map 1 Interface to perceptual data Influenced by the failsafe states set by the navigator Input: Sensory data Output: A map, indicating the feasibility of certain paths for the motion planner Contains: Infeasible regions Restricted regions High cost regions 37

38 MIT Talos Drivability Map 2 38

39 MIT Talos Control System Architecture 39

40 MIT Talos Motion Planner Short term path planning Input: RNDF goals and situational data from the Navigator, Drivability Map Output: Path and Speed commands for the Controller Output is sent at 10 Hz Rapidly-exploring Random Tree Generate semi-random waypoints Iterate over those waypoints Generate a trajectory using closed-loop dynamics Check the trajectory for feasibility 40

41 MIT Talos Control System Architecture 41

42 MIT Talos - Controller Controls the vehicle Generates gas, brake, steering and gearshift commands Two Controllers Pure-Pursuit controller for steering Two different controllers for forward and reverse steering Proportional-Integral controller for speed Steering lookahead is based on current commanded speed Commanded speed is based on vehicle location 42

43 References Leonard, J., How, J., Teller, S. et al. A perception-driven autonomous urban vehicle. In: Journal of Field Robotics, 25 (2008) Nr. 10, p DARPA, DARPA Urban Challenge Website, (2007) Team MIT (2007), Technical Report DARPA Urban Challenge, (2007) Stenzel, R.: Steuerungsarchitekturen für autonome mobile Roboter, Aachen, RWTH, Dissertation, Wikipedia: DARPA Grand Challenge. (2009, December 30) Wikipedia: Autonome mobile Roboter. (2009, November 21) Wikipedia: VaMP. (2009, December 14) ARGO Project Homepage, Univ.-Prof. Dr.-Ing. Ernst Dieter Dickmanns, Forschungsbericht bis , (2002) 43

44 Thank you for your attention! 43