170 likes | 642 Views
For Regular Seminar Robot Software Dec. 23, 2003 Jaesoo Lee jslee@redwood.snu.ac.kr RTOS Lab., SoEECS, SNU Contents Requirements of Robot Software History of Robot Software Architecture SPA (Sense-Plan-Act) Subsumption Hybrid Our Robot Software Architecture
E N D
For Regular Seminar Robot Software Dec. 23, 2003 Jaesoo Lee jslee@redwood.snu.ac.kr RTOS Lab., SoEECS, SNU
Contents • Requirements of Robot Software • History of Robot Software Architecture • SPA (Sense-Plan-Act) • Subsumption • Hybrid • Our Robot Software Architecture • Unexplored Issues Requiring Further Research • Remarks on Robot Research
Robot Software - Requirements • Basic capabilities: deliberation + reactivity • Requirements • Programmability – easiness of introducing new goals • Autonomy and adaptability – aware current goal, execution context • Consistent behavior – reactions guided by objectives • Robustness – fault tolerance • Extensibility – pluggable, reusable, and reconfigurable components • Coordination – multiple goals, multiple (redundant) sensors
SPA (Sense-Plan-Act) • Features • Unidirectional and linear flow of control • Easy execution of a plan (task execution) • Partial orderings, conditionals, and loops • Shortcomings • Planning and world modeling are very hard problems • Open-loop plan execution is inadequate in uncertain and unpredictable environment • Very slow !!
Subsumption • Vertical (hierarchical) decomposition of tasks • Behaviors in higher layer implement more complex, specified task • Each behavior still follows SPA architecture • “The World is its own best model” • No internal models, sense the surroundings on need instead • Direct, ego-centric, and distributed perception • Lisp-like languages to describe behaviors (Augmented Finite State Machine and connecting wires)
Subsumption Level 2: explore Level 1: wander every 10sec Level 0: avoid objects reset
Subsumption • Advantages • Good reactiveness, inherent concurrency • The lower levels have no awareness of higher levels • Competitive coordination with inhibition and suppression mechanisms • Provides a development architecture to incrementally build and test a complex system • Shortcomings • Bad modularity • Upper layers interfere-with/depend-on the internal functions of lower-level behaviors • Increasingly complex • Bad runtime flexibility • Priorities are hardwired • No support for reverse suppression • No global planning
Hybrid Architecture • Goal: reactive planning • Slow planning (SPA, vulnerable to stale decision) • + fast reactivity (Subsumption, subject to fail on sensor errors) • Approach: reintroducing/separating planning • Organization: Plan, (“Sequencing”), Sense-Act
Hybrid Architecture • 3 layer architecture • Deliberate/planning/deliberator • State reflecting predictions about the future • Reasoning, deliberation, planning, world modeling • Task-execution/sequencing/sequencer • State reflecting memories about the past • Coordination of multiple tasks • Reactive/skill/controller/behavior • No state, relying on current knowledge • Tight feedback control loops Ex. NASA 3T Architecture
Huddles and The Trends • Planning, world modeling and increasing complexity of autonomous mobile robot system are turned out to be huddles very hard to solve. • Among these, AI and robot people are trying to solve the complexity problem by eliminating direct communications between modules • TCA by Reid Simmons at CMU • Central Control Module, one of framework component, routes messages, manages resources, and executes task sequences • CORBA is getting used for telerobotics
Our Approach • Applications in our view point • Implements task to achieve specific goals • By concatenating behaviors using partial orderings, parallel executions, conditionals and loops • Has activation conditions, parameter lists, and affinity
Robot software component architecture Componentization Modularity, reusability ORB middleware & component model Communications Modularity Robot software component communication architecture Deployment, implementation Transparency, efficiency, reusability, retargetability, time-to-market UML-based development tool Integrated robot simulator Design, implementation, simulation, packaging Our Approach
Unexplored Issues • Middleware issues on QoS • End-to-end QoS (deadline, delay, jitter, throughput) • Global priority • Reservation-based vs. priority-based communication • Collocated optimization • Schedulability analysis based on constraints annotation • Control quality • Clock synchronization • Real-time event, time-triggered event
Unexplored Issues • Architectural issues • Robot software component communication architecture • Prioritized suppression/inhibition • Subscription/Un-subscription of sensor data • Application affinities • Sequencer service • Reconfiguration on-the-fly and permanent storage • Accustom the robot • Affinities and default parameters for applications • Learning results, acquired landmarks, constructed maps • Requires corresponding supports of underlying ORB, OS, and hardware
Remarks on Robot Research • We can do constructing framework better than them, but how about planning and controls ? • Even though our approach seems to be on the right way, our potential contributions would be small compared to the paid efforts