Aerial robots that interact with the environment

Transcription

1 Aerial robots that interact with the environment Guillermo Heredia*, Aníbal Ollero * Professor at University of Seville, Spain Robotics, Vision and Control group (GRVC) guiller@us.es Robotics, Vision and Control group 1

2 Outline Physical interaction of aerial robots Aerial manipulation Modelling and control Coordinated control Experiments Conclusions Robotics, Vision and Control group 2

3 Introduction UAVs/Aerial Robots/RPAS/Drones: great increase in applications in last years. Most applications use UAVs as observation platforms that carry sensors/communication equipment. Next step: aerial robots that not only observe, but physically interact with the environment. Aerial manipulation: joints mobility of aerial platform and dexterity of manipulator arm. Opens new applications like: Structure assembly at inaccessible sites. Assembly of machines/antennas. Maintenance of industrial plants, aerial power lines. Robotics, Vision and Control group 3

4 Physical interaction of aerial robots Interaction: reciprocal action or influence Physical interaction of aerial robots: Air-to-air refuelling. Cargo delivery/deployment. Multiple UAVs transporting a load. Tethered helicopter. Aerial robot for soil sampling. Air-to-ground interaction: contact. Air-to-ground interaction: Manipulation. Robotics, Vision and Control group 4

5 Air-to-Air Refuelling Physical link between two UAVs: boom or hose. Aerodynamic interaction: tanker stream on receiver. Autonomous refuelling. SAVIER project Robotics, Vision and Control group 5

6 Transportation / Delivery Recent plans for cargo delivery: Amazon/Google. Africa Flying Donkeys. Robotics, Vision and Control group 6

7 Multiple UAVs connected to load Autonomous joint load transportation and deployment. Increase payload. Interaction between the 3 UAVs and load (forces/ torques). Experiments in FP6 AWARE project ( ), DLR. Robotics, Vision and Control group 7

8 Physical interactions: Tethered Helicopter Helicopter linked to ground through a tether. Used as an aid for landing on ships in bad weather. In EC-SAFEMOBIL FP7 project. Robotics, Vision and Control group 8

9 Physical interactions: Tethered Helicopter Robotics, Vision and Control group 9

10 Physical interactions: Tethered Helicopter Robotics, Vision and Control group 10

11 Tethered Helicopter: control approach Robotics, Vision and Control group 11

12 Tethered Helicopter: concept demonstration EC-SAFEMOBIL experiments (CATEC, DLR, Univ. of Seville). Robotics, Vision and Control group 12

13 Aerial robot for soil sampling UGAV: Unmanned Ground Aerial Vehicle, for long range aerial/ground missions. Fixed-wing aerial robot: 3 m wingspan, pusher propeller, twin boom configuration. 5 HP gasoline engine and 1 h. endurance. Autopilot for autonomous operation, attitude/ navigation sensors (IMU, GPS, airspeed), long range radio modem, cameras. Ground robot: four wheels skid-steering electric vehicle with large all-terrain wheels, arm with a shovel for ground samples. Autonomous controller, GPS and orientation sensors, radio link with aerial robot. Elevation mechanism for deployment/ docking of ground robot. Robotics, Vision and Control group 13

14 UGAV experiments Robotics, Vision and Control group 14

15 Air-to-ground interaction Force interaction: contact. Contact inspection (i.e. ultrasounds, eddy current) Cleaning with special devices Manipulation: Robotic manipulation with multi-joint arms Robotics, Vision and Control group 15

16 ARCAS project Robotics, Vision and Control group 16

17 ARCAS FP7 project: The ARCAS project aims at the research, development and experimental validation of the first cooperative flying robots system for assembly and manipulation. Technical objectives: Flying robots with arms and motion control systems, including coordinated control of several flying robots Aerial perception: scene recognition, range-only SLAM, 3D tracking, accurate positioning and cooperative perception Aerial Cooperative assembly planning Robotics, Vision and Control group 17

18 New platforms developed in ARCAS Helicopters with 7-dof arms (DLR) Multirotor with 6-dof arm (CATEC) Multirotor with 7-dof arm (USE) Robotics, Vision and Control group 18

19 Design of new 6 dof arm (CATEC) Robotics, Vision and Control group 19

20 USE outdoor multirotor Development of multirotor for outdoors (realistic scenarios). Outdoor aerial robot positioning: needs additional sensors for positioning. Accurate positioning sensor (RTK-DGPS). Ultrasonic sensor for multirotor/ground relative height. Vision sensors with onboard processing for: Aid for multirotor positioning. Arm end-effector accurate relative positioning (object manipulation). Manipulator arm: greater reach (larger workspace). Multirotor needs larger payload to accommodate sensors for outdoor operation and greater reach arm. Robotics, Vision and Control group 20

21 AMUSE multirotor AMUSE: Aerial Manipulator from USE. Coaxial-quad multirotor configuration: four arms (82 cm diameter) with two coaxial rotors in each arm. Eight 750 W motors with 16 rotors. Payload for high level computer, sensors, arm and objects: 5 kg. AMUSE Robotics, Vision and Control group 21

22 AMUSE manipulator arm Arm: Cyton Gamma 1500 from Robai. 7 Degrees of freedom, modified Dynamixel servos (compatible interface). Maximum reach: 0,86 m. vertical; 0,52 m lateral. Weight: ~ 2 kg Arm workspace Robotics, Vision and Control group 22

23 Tensor-driven lightweight hand Design and development of a tensor-driven, spring loaded lightweight hand with three fingers: wider opening (less accurate positioning needed), better object grasping. Robotics, Vision and Control group 23

24 AMUSE electronics CATEC autopilot. IMU. Barometer. High Level Computer: intel i7 Camera 1 looking downwards. Camera 2 looking to arm workspace. Camera 3 at end effector (not mounted). 7 dof manipulator arm connected to autopilot autopilot integrated controller (multirotor + arm). Robotics, Vision and Control group 24

25 25 Modelling and control of multirotor with manipulator arm Usual assumptions in standard multirotor modelling and control: Rigid body. NOT VALID Symmetrical. Main effects that modify the dynamic behavior of the multirotor with a manipulator with respect to a standard multirotor configuration: 1. Displacement of the center of mass from the vertical axis at the geometrical center of the multirotor. 2. Variation of mass distribution. Variation of moments of inertia. 3. The dynamic reaction forces and torques generated by the movement of the arm. Robotics, Vision and Control group 25

26 Effects of arm on multirotor control Torque generated by displacement of center of mass. 5 4 TORQUE [N*m] JOINT-2 [DEG] Robotics, Vision and Control group 26

27 Effects of arm on multirotor control Variation of moments of inertia I xx Ixx [Kg*m 2 ] JOINT [DEG] I xz Ixz [Kg*m 2 ] JOINT [DEG] Robotics, Vision and Control group 27

28 Effects of arm on multirotor control Dynamic reaction torque in base generated by movement of the arm. 8 6 TORQUE [N*m] TIME [seg] Robotics, Vision and Control group 28

29 29 Modelling and control of multirotor with manipulator arm Multirotors: underactuated systems, subject to perturbations (wind/turbulence): difficult to maintain position/attitude. Manipulator arm: fully actuated, use servos with embedded controllers, easy position control (with actual arms mounted on ARCAS multicopters). Full dynamic model of quadrotor with arm: complex, coupled nonlinear dynamics. Can be derived using Euler-Lagrange: with, : generalized state, u: actuator forces/torques, u ext : external forces/torques, B: mass matrix, C: centrifugal/coriolis terms, g: gravity terms. Robotics, Vision and Control group 29

30 30 Multirotor controller Full 3D multicopter+arm dynamic model considered for controller derivation., Full-Dynamics Integral Backstepping (FD-IB) nonlinear controller for the attitude angles.,, ] is a vector with the roll, pitch and yaw control inputs, the controller terms can be arranged in the following matrix form:,, Robotics, Vision and Control group 30

31 31 Multirotor controller Full 3D multicopter+arm dynamic model considered for controller derivation., Full-Dynamics Integral Backstepping (FD-IB) nonlinear controller for the attitude angles.,, ] is a vector with the roll, pitch and yaw control inputs, the controller terms can be arranged in the following matrix form:,, Standard PID controller. is a vector with angular position errors for roll, pitch and yaw, ( : angular velocity errors, : integral of angular position error)., and are diagonal matrices with the parameters. Robotics, Vision and Control group 31

32 32 Multirotor controller Full 3D multicopter+arm dynamic model considered for controller derivation., Full-Dynamics Integral Backstepping (FD-IB) nonlinear controller for the attitude angles.,, ] is a vector with the roll, pitch and yaw control inputs, the controller terms can be arranged in the following matrix form:,, Variable Gain matrix. Depends on the moments of inertia of the robot. Varies when the arm moves. It also includes tuning parameters. Robotics, Vision and Control group 32

33 33 Multirotor controller Full 3D multicopter+arm dynamic model considered for controller derivation., Full-Dynamics Integral Backstepping (FD-IB) nonlinear controller for the attitude angles.,, ] is a vector with the roll, pitch and yaw control inputs, the controller terms can be arranged in the following matrix form:,, Gravity compensation term. Static compensation of the torque caused by displacement of the center of mass from the vertical of the geometric center of the multirotor. It depends on the angular positions of the arm joints. Robotics, Vision and Control group 33

34 34 Multirotor controller Full 3D multicopter+arm dynamic model considered for controller derivation., Full-Dynamics Integral Backstepping (FD-IB) nonlinear controller for the attitude angles.,, ] is a vector with the roll, pitch and yaw control inputs, the controller terms can be arranged in the following matrix form:,, Dynamic torque compensation. Dynamic and Coriolis terms, compensate the dynamic torques generated when the arm moves. Robotics, Vision and Control group 34

35 35 Multirotor controller Full 3D multicopter+arm dynamic model considered for controller derivation., Full-Dynamics Integral Backstepping (FD-IB) nonlinear controller for the attitude angles.,, ] is a vector with the roll, pitch and yaw control inputs, the controller terms can be arranged in the following matrix form:,, In this form, the FD-IB controller can be adapted and tuned starting from standard PID-based baseline multirotor controllers FD-IB position controllers can be derived in a similar way. Robotics, Vision and Control group 35

36 36 FD-IB multirotor controller simulation Simulation of multirotor maintaining hover. From t = 15 s. to 17 s.: wide swing of arm. Comparison with standard cascaded PID controller (not considering arm). 8 ATTITUDE ANGLES ROLL (º) 0-2 PITCH (º) PID -FDIB PID -FDIB Time (s) Time (s) Robotics, Vision and Control group 36

37 37 FD-IB multirotor controller simulation Simulation of multirotor maintaining hover. From t = 15 s. to 17 s.: wide swing of arm. Comparison with standard cascaded PID controller (not considering arm) POSITION X, Y, Z X (m) Y (m) Z (m) PID -FDIB PID - FDIB PID -FDIB Time (s) Time (s) Time (s) Robotics, Vision and Control group 37

38 38 FD-IB attitude control experiments Experiments with AMUSE multirotor with 7 dof arm. Multirotor in hover, command large excursion movements to arm (worst case, large variations of mass Pitch angle 10 center and inertias). PID 5 Comparison of FD-IB with standard PID : oscillations with PID almost 0 double FD-IB. -5 Remaining oscillations due to wind Time [s] and position controller. PITCH [deg] 10 5 FD-IB PITCH [deg] Time [s] Robotics, Vision and Control group 38

39 39 Arm controller Hardware restrictions for arm controller: use of Dynamixel or standard servos for arm joint actuation. Difficult to use torque input. Implementation of admittance controller for contact tasks: command a desired cartesian position for arm end effector Σ : Σ Σ Σ Σ : desired cartesian position of Tool Center Point (TCP). Σ : additional displacement that would get the desired interaction forces and torques between end-effector and objects/environment. Then, Σ is transformed through the manipulator inverse kinematics. Desired joint position setpoints are transmitted to servos. Arm inverse kinematics : Jacobian-based first-order algorithm. Redundant 7-DoF arm motion: generated through jacobian null space. Arm extra DoF: maximize distance from mechanical joint limits. Robust behavior close to singular configurations: modified pseudoinverse with variable damping factor based on gaussian-weighted functions of the manipulability measure. Robotics, Vision and Control group 39

40 40 Arm control experiments Experiments with arm following references from video system: Blue: joint references computed by arm controller. Green: joint trajectories. End-effector position and attitude errors Position Errors [m] Orientation Errors [deg] X error Y error Z error error error error Time [s] Manipulator s Joint Angles [deg] Robotics, Vision and Control group Time [s]

41 Aerial robots that interact with the environment ARCAS experiments Robotics, Vision and Control group 41

42 Coordinated Control (UNINA/UNICAS/UNIBAS) Robotics, Vision and Control group 42

43 Coordinated Control 4 th year experiments Robotics, Vision and Control group 43

44 What s next? Recently funded H2020 project AEROARMS: AErial RObotic system integrating multiple ARMS and advanced manipulation capabilities for inspection and maintenance Robotics, Vision and Control group 44

45 Conclusions First steps of cooperative aerial manipulation First world-wide demonstrations: aerial robots general manipulation with multi-joint arms, structure assembly. Control approach for aerial robots with multilink manipulators which considers full dynamics. Tested in simulation and experiments. First outdoor experiments with Aerial Manipulator. Large number of applications Intensive experimentation is needed Small systems: easier to operate (indoors) but limited applications (operational constraints). Larger payload systems are required for many applications (outdoors). Robotics, Vision and Control group 45

46 Aerial robots that interact with the environment Guillermo Heredia*, Aníbal Ollero * Professor at University of Seville Robotics, Vision and Control group (GRVC) guiller@us.es Robotics, Vision and Control group 46