LAUROPE Six Legged Walking Robot for Planetary Exploration participating in the SpaceBot Cup

Transcription

1 FZI RESEARCH CENTER FOR INFORMATION TECHNOLOGY LAUROPE Six Legged Walking Robot for Planetary Exploration participating in the SpaceBot Cup ASTRA 2015, ESA/ESTEC, Noordwijk 12. Mai 2015

2 Outline SpaceBot Cup Challenge LAURON V & Competition setup PlexNav Framework Localisation, Mapping and Navigation Object detection and Tracking Mobile Manipulation Lessons learned & Conclusion FZI Forschungszentrum Informatik 2

3 Outline SpaceBot Cup Challenge LAURON V & Competition setup PlexNav Framework Localisation, Mapping and Navigation Object detection and Tracking Mobile Manipulation Lessons learned & Conclusion FZI Forschungszentrum Informatik 3

4 SpaceBot Cup - Tasks Planetary Exploration and Sample Return mission The challenge: Navigate in previously unknown Terrain Map the environment Find 3 predefined objects Collect 2 and assemble them at a third Return to start (with obstacles) Additional obstacles Control only via ground station Time delay on communication (2s in each direction) Bidirectional communication only at Checkpoints High level of autonomy required FZI Forschungszentrum Informatik 4

5 SpaceBot Cup - Environment Mostly sand and rocks Gravel Soft sand (deep) Large boulders Various inclines Minimum 15% Maximum 30% Overlaps Obstacles in the obvious path Object positions unknown Rough height map given shortly before the competition High level of mobility required FZI Forschungszentrum Informatik 5

6 LAUROPE LAURON V 6 legged bio inspired walking robot as main system Technical data: Footprint: 90 cm x 80 cm Weight: about. 42 kg 4 DoF per Leg, 2 DoF in the Head 26 Degrees of Freedom Modular Sensorhead (Stereo, IR Camera, LIDAR, ) Behaviour Based control Battery runtime > 120min. Behaviour Based control allows traversal of unknown terrain (without sensor information) Delta joint allows adaption to steep inclines and the use of the front legs for mobile manipulation FZI Forschungszentrum Informatik 6

7 LAUROPE - LAURON V Posture behaviours govern the overall robot pose Reactive leg behaviours adapt to the environment FZI Forschungszentrum Informatik 7

8 LAUROPE LAURON V KaRoLa Karlsruhe Rotating Laserscanner Provides high resolution (~ points) 3D Pointcloud of the environment Custom made mandible gripper Specifically designed for walking robots Can lift up to 2,3kg Objects of up to 50cm diameter Object containers Mounted based on kinematic analysis Various modifications Chains instead of belts Larger feet for softer sand FZI Forschungszentrum Informatik 8

9 Outline SpaceBot Cup Challenge LAURON V & Competition setup PlexNav Framework Localisation, Mapping and Navigation Object detection and Tracking Mobile Manipulation Lessons learned & Conclusion FZI Forschungszentrum Informatik 9

10 Planetary Exploration and Navigation Framework PlexNav Architecture PlexNav Data Mission Layer Goal List Mission Control Skill Layer Capability Management Task Allocation Object List Perception Manipulation Special Sensors Environment Global Map Local Maps Robot Layer Behaviour Control Robot Abstaction Hardware Abstraction Simulation Interface FZI Forschungszentrum Informatik 11 ROADS

11 Skills Localisation & Mapping Visual odometry (FOVIS) based on RGBD data very accurate localisation but can drift/fail 3D Laserscans taken by KaRoLa Pointcloud matching with ICP and VO as prior Poses are saved as factor graph Feature based loop closing Individual and overall map are used for PlexMaps FZI Forschungszentrum Informatik 12

12 Skills - Planning PlexMap as basis for planning and visualisation 2,5 D Height map Quadtree data structure ( based on Octomap) Each cell stores height, variance (roughness) and resolution Modified RRT* to generate plans that respect the robots capabilities (considering roughness, available data, inclination and footprint inclination) Successfully transferred the approach to KAIRO FZI Forschungszentrum Informatik 14

13 Skills Object detection & tracking Simple yet effective color based detection Pixel blob detection Size and color evaluation (confidence) Tracking and position estimation Verification Works with on RGB only 3D pose estimation with fitting of geometric shapes FZI Forschungszentrum Informatik 15

14 Skills Object detection & tracking FZI Forschungszentrum Informatik 16

15 Skills Object manipulation Custom gripper can grasp both target objects in the same gripping pose Current measurements provide feedback Predefined manipulation pose (restricted kinematic) and storage procedures Final grasping position generated from object fitting (perception) Body is used to compensate underactuated kinematics FZI Forschungszentrum Informatik 17

16 Skills Object manipulation FZI Forschungszentrum Informatik 18

17 Outline SpaceBot Cup Challenge LAURON V & Competition setup PlexNav Framework Localisation, Mapping and Navigation Object detection and Tracking Mobile Manipulation Lessons learned & Conclusion FZI Forschungszentrum Informatik 19

18 Lessons learned during the Competition LAUROPEs run in the competition was not successful Introspection for autonomous systems is critical Main reason for the SBC 2013 to not work properly was the missing information which system actually failed (mission control) Autonomy is key The delayed communication effectively prohibited any ground station control Without the capability to intervene, the system needs to be designed with a top level approach Rather than individual skills, the overall system should be the focus Flexible mission control for activation of various skills is required FZI Forschungszentrum Informatik 20

19 Lessons learned during the Competition Simulations and early analogue tests are very valuable Loop closing feature set could be used effectively and was trained completely in simulation Self build objects and delayed communication allowed to test these subsystems early on ROS as framework worked very well for high level tasks Loosely coupled structure allowed to quickly adapt during development Visualisation tools already provided are very useful Walking robots for space exploration Six legs can handle challenging terrain Mobile manipulation with the legs is very feasible FZI Forschungszentrum Informatik 21

20 Conclusions LAUROPE Bio inspired hexapod LAURON V as robot for the Planetary exploration Behaviour based control enables terrain adaption Prepared for rough terrain and mobile manipulation Can overcome even steep slopes without sensor data KaRoLa for localisation and mapping PlexNav High level framework for planetary exploration Provides skills like localisation, mapping and object detection Robot agnostic implementation allows adaption to other robots Full 3D localisation and accurate map building even in sparse environments SpaceBot Cup Walking robots can be used quite well in such an environment and to fulfil the required tasks FZI Forschungszentrum Informatik 22

21 FZI Forschungszentrum Informatik 23

22 Questions? FZI Forschungszentrum Informatik 24

23 PlexNav Architecture Data Mission Layer Goal List Mission Control Skill Layer Capability Management Task Allocation Object List Perception Manipulation Special Sensors Environment Global Map Local Maps Robot Layer Behaviour Control Robot Abstaction Hardware Abstraction Simulation Interface FZI Forschungszentrum Informatik 25

24 PlexNav - Mission Control State Monitoring & Update Ground Control Capability Finished/ Aborted Task Monitoring & Command Publish current State & Execute next Capability Task Management Selection of Capability {id1,id2,string} Goal List Object List Capabilities Fine Localize Object Refine Manipulation Pose Move to POI Manipulate FZI Forschungszentrum Informatik 26

25 PlexNav - Mission Control Execution of individual capability is governed by a state machine provided with it -> complexitiy of the interface is reduced MoveToPOI Skill Fine Localization Skill FZI Forschungszentrum Informatik 27

26 PlexNav - Communication Uses available ROS tools Diagnostic functionality for monitoring Permanent monitoring of mission status Communication with ROS bridge master connect ( plexmap, TCP) TCP server: :2342 connect( :2342) Plexmap data messages FZI Forschungszentrum Informatik 28

27 PlexNav - Communication Uses available ROS tools Diagnostic functionality for monitoring Permanent monitoring of mission status Communication with ROS bridge master master FZI Forschungszentrum Informatik 29

28 LAUROPE Early test systems Sensor testbed for selection of appropriate sensors Omnidirectional platform with final sensor setup for tests of individual skills and functions Gazebo based simulation environment for development of walking capabilities and overall mission ROS Based software for easy integration FZI Forschungszentrum Informatik 30