Autonomous Mission Planning And Execution for Two Collaborative Mars Rovers

Transcription

1 Autonomous Mission Planning And Execution for Two Collaborative Mars Rovers Juan Manuel Delfa Brian Yeomans Yang Gao, Oskar von Stryk ASTRA 13/05/2015

2 Introduction Lot of work conducted in path planning and mobility. Other types of autonomy use to be forgotten. NASA has developed several autonomous systems for MER and Curiosity: On-Ground: MAPGEN mission planner. On-Board: AEGIS opportunist science, Autoplace, VTT, WATCH. This presentation talks about tactic/strategic Mission Planning and Execution.

3 Motivation: More Autonomy Automated Mission Planning Process that, given and initial and goal state, chooses and organizes actions by anticipating their expected outcomes in order to achieve the goals. Automated Execution Process that, given a plan composed of sequences of activities, command each activity to the appropriate subsystem at the appropriate time. Problem: Temporal Planning for complex problems is not solvable with modern computers.

4 The FASTER project FASTER (Forward Acquisition of Soil and Terrain data for Exploration Rover) EU-FP7 Program funded by European Union. Consortium including. Goal: To improve the security and speed of planetary rovers. Mission concept: Two collaborative rovers working in tandem: Primary: ExoMars-class. Scout: Smaller, more robust, travering in front to detect potential hazards.

5 FASTER - Rovers Primary Rover Bridget: ExoMars testbed. Constructed by Airbus. Modified with extra-sensors for soil-sensing. Scout Rover High mobility. Constructed by DFKI and UoS.

6 Operations cycle for a Rover Operations divided in two groups: On-Ground On-Board Definition of the plan goals. Definition of the Knowledge DataBase (Behaviours). Plan generation. Uplink of the plan. Execution. Plan repair.

7 Architecture Adaptable level of autonomy (E1- E5). Adaptable internal organization for different mission requirements. Thee-layer architecture Deliberative: Long-term planning. Reactive: Execution and short-term repair. Functional: Hardware dependent controllers.

8 Architecture - Components Deliberative: RobCon: Middleware between Planner and Executive. Mission Planner: QuijoteExpress. Reactive: Primary rover Executives: SanchoExpress. Scout rover: Specific executive (DFKI) subordinated to SanchoExpress. Functional: Soil Sensor System. Plath Planner. Bridget Locomotion System.

9 Mission Planner - QuijoteExpress Behaviours Planner Solution Model Problem

10 QuijoteExpress - Novelties Provide high levels of autonomy and collaborative capabilities to robots Properties 1. How to produce faster solutions Parallel planning Forward-chaining planning Heuristic guided 2. How to manage unknown situations Partial planning 3. How to better help engineers to verify & validate plans Hierarchical models & behaviours High level goals

11 QE FASTER Modelling Constraint Action Subsystem

12 QE Traverse Behaviours

13 QE Traverse Behaviours

14 From Mission Planning to Execution Plan is represented with Timelines. Timeline: Sequence of actions with an starting and ending time frame. One timeline per subsystem. Transition: Time at which at least one timeline change value.

15 SanchoExpress: Properties Provides autonomous execution for robotic systems Properties 1. Temporal Execution Flexible timelines 2. Parallel execution All timelines executed in a dedicated thread 3. FDIR Capabilities Primary Locomotion can detect and fix failures derived from traverse activities.

16 SE: Dedicated Executives Generic Executive Platform independent. Receives a plan. Commands individual actions. Reports back to deliberative layer (user or planner depending on level of autonomy). Decides when to execute (next transition). Dedicated Executive Platform dependent. One for each dedicated subsystem. Receives/commands individual actions. Reports back to Generic Executive. FDIR capabilities.

17 SE: Example 1. Generic Executive (GE) receives new action TurnToWP. 2. Searches in a DataBase which dedicated executive knows how to execute it. 3. Primary Locomotion (PRLoc) calls the method TurnToWP, which contains 3 actions. TurnToWP(goal) Why this approach? No need to plan in such low level of detail. User doesn t need to know unless something wrong happens.

18 FASTER Tests & Final Demo Multiple test-campaigns: 5 Integration: Oriented to integrate all subsystems. 5 Mission-level: Oriented to make the subsystems work together. Testing Mission Planner and Executive Specially difficult: Both depend on the rest of subsystems. Simulation campaigns with dummy subsystems. Final Demonstration: Presented to European Commission on October, 2014 in the Mars Yard facility, Airbus DS, Stevenage.

19 FASTER Setup Rovers in formation Rocks of different sizes Loose soil Sand trap

20 FASTER Plan Flexible plans generated in few seconds. Final demo mission: Traverse 15 meters. Demonstrate Soil Sensor System. Estimated 12 traverse cycles. Each traverse cycle containing 18 transitions. 5 timelines.

21 FASTER Workflow Choose Target Planning Generate Map Generate Path Primary Traverse Update map Scout Traverse

22 Results Mission Planner Great performance. Time required negligible compared with more intensive processes related to image processing. Hierarchical planning worked successfully. Flexible plans were crucial for the success. Models and behaviours worked successfully. Executive Worked flawlessly. During final demo, short-scope plan repair was required twice, helping to resume the mission without the need of re-planning.

23 Conclusions Several new technologies has been demonstrated Novel mission planner. Generation and Execution of collaborative plans. On-the-fly plan repair. Next steps Achieve opportunistic science. Demonstrate re-planning with next generation QuijoteExpress. Go further in the integration with the path planner.

24 Questions Contact