Massachusetts Institute of Technology (Videos not included in the repository) Task Level Collaboration with Robotic Assistants “Progress Towards Task-Level Collaboration Between Astronauts and Their Robotic ATRV Rover Testbed Humanoid Simulators Assistants” Robert Effinger, Andreas Hofmann, Prof. Brian Williams Model-Based Embedded and Robotic Systems Group Massachusetts Institute of Technology These Robotic Assistants must be able to: ISAIRAS 2005 1.) Interpret task level commands 2.) Execute the tasks safely and reliably, even under disturbances and execution uncertainties 1.) Interpreting task level commands 1.) Interpreting task level commands Temporal Plan Network Reactive Model-Based (Kim, Williams, Abrahmson IJCAI 01) A Motivating Example: Lunar Roving Vehicle Deployment Programming Language (Williams et.al., ISAIRAS 01) A [ lb , ub ] A [ lb , ub ] A [ lb , ub ] A [ lb , ub ] A [ lb , ub ] [ lb , ub ] [ lb , ub ] [ lb , ub ] [ lb , ub ] c [ lb , ub ] c [ lb , ub ] Tell (c) Tell (c) Tell (c) RMPL := c | act | TPN1 | … … … TPN1 TPN1 TPN1 TPN2 TPN2 TPN2 TPN1 [ lb , ub ] | sequence ( TPN1 , TPN2 , … ) sequence ( TPN1 , TPN2 , … ) parallel ( TPN1 , TPN2 , … ) | sequence ( TPN1 , TPN2 , … ) | TPN1 TPN1 choose ( TPN1 , TPN2 , … ) | parallel ( TPN1 , TPN2 , … ) parallel ( TPN1 , TPN2 , … ) TPN2 TPN2 if (c) thennext ( TPN1 ) | … … do ( TPN1 ) maintaining (c) TPN1 TPN1 TPN1 choose ( TPN1 , TPN2 , … ) choose ( TPN1 , TPN2 , … ) TPN2 TPN2 TPN2 … … … Ask (c) Ask (c) Ask (c) TPN1 TPN1 TPN1 if (c) thennext ( TPN1 ) if (c) thennext ( TPN1 ) (*picture courtesy of NASA) TPN1 TPN1 TPN1 do ( TPN1 ) maintaining (c) do ( TPN1 ) maintaining (c) Ask (c) Ask (c) Ask (c) Apollo LRV Deployment Sequence Abstract Task Decomposition Deploy Seats and Footrests LRV-deployment-sequence() [5,20] = { sequence( [5,10] R: Remove insulation blanket [1,3], R: Remove operating tapes [1,3] parallel( R: Deploy seats A: Lower LRV w/braked reel [1,5], and footrests do ( sequence( RMPL R: deploy aft wheels [0.5,2], [3,7] [2,3] R: deploy front wheels [0.5,2] ) R: Deploy R: Deploy )maintaining (tension on cable) seats footrests ) choose( A: deploy seats & footrests [1,5], R: deploy seats & footrests [5,10] [1,2] [1,3] [1,2] ) ) } [5,20] R: Deploy right R: Walk to left side R: Deploy left seat of LRV seat TPN (*picture courtesy of NASA) [1,5] [1,5] A: Lower LRV using [1,3] [1,3] A: Deploy seats braked reel and footrests R: Remove R: Remove [0.5,2] Insulation Operating [0.5,2] [5,10] Tapes Blanket R: Deploy R: Deploy R: Deploy seats aft wheels front wheels and footrests (*picture courtesy of NASA) Ask( tension on deploy cable) 1
Task Level Collaboration with Robotic Assistants Flexible Execution Times (Dechter, Meiri, Pearl 91) ATRV Rover Testbed Humanoid Simulators Simple Temporal Network (STN): [ ] − ∈ [1,5] T T 1 , 5 End- Begin- j i engine-start engine-start Equivalent Distance Graph Representation: u [l,u] A B A B These Robotic Assistants must be able to: -l 1.) Interpret task level commands STN Distance Graph [ ] − ∈ ⇒ − ≤ ∩ − ≤ − 2.) Execute the tasks safely and reliably, even under T T l , u T T u T T l j i j i i j disturbances and execution uncertainties Flexible Execution Times Flexible Execution Times (Dechter, Meiri, Pearl 91) (Dechter, Meiri, Pearl 91) A Simple STN: A Simple STN: 5 1 [1,5] Consistent ! [5,1] Inconsistent STN ! A B A B A B A B d = 0 d = 1 d = 0 d = 5 d = - 4 d = - 8 d = - 3 -1 -5 Determine STN consistency: Determine STN consistency: - Calculate the Single Source Shortest Path (polynomial-time algorithm) - Calculate the Single Source Shortest Path (polynomial-time algorithm) - A continually looping negative cycle indicates an inconsistency in STN Two methods to detect a continually looping negative cycle (most space efficient) 1.) Check for any d-value to drop below –nC. 2.) Keep an acyclic spanning tree of support, and terminate when a self-loop is formed. (Cesta, Oddi 96) (most time efficient) Incremental Reasoning Algorithm Performance Improvements UAV Scenarios • Basic Idea: Legend: Comparison of Algorithm Runtime Comparison of Algorithm Runtime Seeker UAV NF NF Water UAV Algorithm Runtime (sec ) Algorithm Runtime (sec) 12 12 Z1 1.) Keep dependency information for each shortest-path value in No-Fly Zone Z2 10 10 WaterB UAV Fire2 Fire Non-incremental Non - incremental Base 8 8 the distance graph (Cesta, Oddi 96) Water Algorithm Algorithm UAV Base 6 6 Incremental Incremental Algorithm Algorithm 4 4 Fire1 WaterA 2 2 Plan Goal: Extinguish All Fires 0 0 2.) Use incremental update rules to localize necessary changes to the Vehicles: Two Seeker UAVs 1 1 10 10 19 19 28 28 37 37 46 46 55 55 64 64 73 73 82 82 91 91 One Water UAV Number of Activities Number of Activities distance graph. Resources: Fuel & Water a.) 3 Update Rules to change a consistent distance graph. Randomly Generated Plans b.) 3 Update Rules to repair an inconsistent distance graph. • ITC’s Novel Claims: 1.) A conflict extraction mechanism to guide plan repair 2.) Allow multiple arc-changes 3.) Can repair inconsistent distance graphs incrementally 2
Robustness To Environmental Disturbances Qualitative Behavior Specification for Locomotion and Uncertainty – Andreas Hofmann, PhD Thesis Gait Poses [5,10] l1 r1 r2 R: Walk to Right Side of LRV [5,10] R: Walk to Right Side l1 l1 r2 r2 l2 of LRV start finish • [Muybridge, 1955] CM ∈ R [t_lb, t_ub] 1 CM • Stop-action left left photographic study toe-off heel-strike of human and Left Foot animal motion right right toe-off heel-strike • Gaits depicted as Right Foot sequences of distinct qualitative Foot placement poses l1 l2 Lat d 2 CM ⎛ ⎞ = − M CP CM ⎜ tot ⎟ ⎝ K ⎠ Fwd r1 r2 dt 2 Flexible spatial and temporal constraints Walking with constrained foot Nominal Walking placement • Allows for linearizing • Implemented in controller controllers that decouple through Lagrangian state variables and makes • Angular momentum tightly relaxation them directly controllable conserved during normal walking • Orientation goals lower • [Hofmann, et al; 2004] priority than balance goals A Dimensionless Spin 0.02 & Medial−lateral & & - y y y y _ 0 ∫ ∫ 1 set 1 1 1 + k + −0.02 p • When disturbed, sacrifice −0.04 & 0 20 40 60 80 100 y _ 1 set Dimensionless Spin Anterior−posterior 0.01 k tight angular momentum + d - 0 conservation temporarily −0.01 0 20 40 60 80 100 – Until balance restored Dimensionless Spin 0.01 Vertical 0 −0.01 0 20 40 60 80 100 Gait Cycle ( % ) Hybrid executive coordinates controllers Plan compilation for efficient execution - to sequence biped through qualitative state plan State Plan State Plan Model-based Model-based Executive Executive Compiler: Plan Compiler • Computes tubes of feasible control Hybrid Task-level trajectories from Qualitative State Plan. Qualitative Executive Lf_1 Control Plan Rf_1 Rf_2 Lf_1 Rf_2 Lf_2 Lf_1 Rf_2 Hybrid Dispatcher Dispatcher: Plant Plant Control Control state state parameters parameters • dynamically searches for optimal SISO SISO Linear Linear Systems Systems control trajectories within tubes. Linearizing Multivariable Linearizing Multivariable Controller Controller • Dispatcher “pulls springs” of each state Plant Plant control Plant Plant control variable. state inputs state inputs MIMO Nonlinear MIMO Nonlinear Plant Plant 3
Recommend
More recommend