Aerial Robots for Construction Vijay Kumar UPS Foundation - PowerPoint PPT Presentation

Workshop on Uncertainty in Automation, ICRA 2011 Aerial Robots for Construction Vijay Kumar UPS Foundation Professor Departments of Mechanical Engineering and Applied Mechanics and Computer and Information Science Member of the GRASP Laboratory and the Graduate Group of Computational Biology University of Pennsylvania Acknowledgements ONR N00014-08-1-0696 (HUNT) ONR Grant N00014-09-1-1051 (SMARTS) ARL W911NF-08-2-0004 (MAST) ONR N00014-09-1-1051 (ANTIDOTE) 1

Q. Lindsey, D. Mellinger, V. Kumar, Construction of Cubic Structure with Teams of Aerial Robots, RSS, LA, June 2011 Dr. Nathan Michael Jonathan Fink Daniel Mellinger Mike Shomin Chris-ne Quen-n Lindsey Frank Shen Kappeyne Ma? Turpin 2

Cooperating Robots and Assembly ABB Kuka Kiva Systems Shimizu 3

Unmanned Air Vehicles UCB Smart bird Aerovironment Black Boeing/ Insitu Scaneagle – 33 lb Widow – 2.12 oz. Gen. Atomics – Predator B – 7,000 lb BAE Systems U. Penn Microstar – 3.0 oz. Piper Stanford DFly Boeing X-45A UCAV – 12,195 lb (est) Astec cub 6 lb AAI Shadow 200 – 328 lb Pelican Allied Aero. LADF 3.8 lb Astec Aerovironment Hummingbird Pointer – 9.6 lb Northrop-Grumman Global Hawk 25,600 lb Bell Eagle Eye – 2,250 lb 0 1 10 100 1,000 10,000 100,000 UAV Weight Micro Mini Tactical Med Alt High Alt / UCAV D. Pines , 2005 4

Assembly Construction  Structured  Unstructured  Mass/Batch  Customized  Outdoor  Indoor  Potentially remote,  Human intervention hostile environments usually always possible  Process tolerance  Process tolerance > 5 mm < 0.1 mm 5

Goal Assembly and Construction of 3-D Structures 6

Goal Assembly and Construction of 3-D Structures This talk … Special Cubic Structures 7

Assembly Primitives P1 P2 P4 P3 8 8

Tolerances and Variation Admissible Product Design Manufacturing Tolerances variation Part, assembly Assembly plan Assembly Process Model Process tolerance Process Successful! Robo8c Automa8on, variation Assembly Robo8cs Model Unsuccessful! 9

Assembly Primitives P1 P2 P4 P3 10 10

Special Cubic Structures Structures consisting of layers/strata  No holes in any 2D stratum y x  No cantilevered sections z x 11

Wavefront Raster (WFR) Algorithm 1: Build any square in the 2-D region 2: while not finished do 3: mark squares immediately connected to already built region 4: for (leftmost column) to (rightmost column) 5: build marked squares in column from bottom to top Wave front 3 Wave front 2 Wave front 1 1 1 3 5 1 3 5 1 2 4 4 2 2 12

Quad Rotors [ Mellinger, Michael and Kumar, ISER 2010 ; Mellinger and Kumar, ICRA 2011 ] 13

Cooperative Grasping and Lifting u ∗ = arg min � f T u { J | Au = w } , J = i Qf i 14 i

Part Bins 15

Gripper !"#$% &#'()*%+, 16 16

17

Force Feedback  Can estimate mass, moments of inertia  Confirm stable prehension Estimated Mass (kg) 0.8 Feel/respond to forces 0.7 0.6 0 20 40 60 80 Time (s) 18

Assembly Modes M1 M3 M2 M5 M4 19 19

Assembly Execute Hover at Yaw Yaw Release and Hover at P 1 trajectory from P 2 Left Right Ascend P 1 to P 2 | ψ error | > ψ max Failed assembly, | ψ error | > ψ max repeat attempt y x ψ 20 20

21

Assembly Errors 1 0.5 0 � 0.05 � 0.04 � 0.03 � 0.02 � 0.01 0 0.01 0.02 0.03 0.04 0.05 x (m) 1 M1 0.5 M4 0 � 0.05 � 0.04 � 0.03 � 0.02 � 0.01 0 0.01 0.02 0.03 0.04 0.05 y (m) 1 0.5 M2 0 � 0.05 � 0.04 � 0.03 � 0.02 � 0.01 0 0.01 0.02 0.03 0.04 0.05 z (m) 1 M5 0.5 z 0 y � 30 � 20 � 10 0 10 20 30 � (deg) M3 x M1 M2 M3 M4 M5 22

23

Assembly Results with Three Robots Simulation Number of Parts 32 34 40 192 Successful Trial 1 32 33 40 Attempts Trial 2 32 34 39 Actual Time 449.6 486.6 588.2 450.7 486.2 587.3 Column retries 5 3 8 5 1 3 Beam retries 4 2 5 5 2 1 Time (in simulation) 443.6 480.4 581.9 2642.0 24 24

Challenges  Distributed assembly 25

Challenges  Distributed assembly  Unstructured environments 26

Challenges  Distributed assembly  Unstructured environments  Part design and payloads 27

Robotic Assembly/Construction Admissible Product Design Manufacturing Tolerances variation Part, assembly Assembly plan Assembly Process Model Process tolerance Process Successful! Robo8c Automa8on, variation Assembly Robo8cs Model 28

Conclusion  Agile, small, aerial robots create new opportunities for robotics  SWAP constraints  Force feedback enables adaptation  Networks enable functionality beyond what can be achieved by individual robots 29