Closing Remarks

Abstract

The work presented in this book constitutes a step towards an integrated framework to address the problem of driving a UAM using visual information, including the robot state estimation and high-level task control laws. In this final chapter we discuss the conclusions derived from each chapter of this book, gathering them into an overall reflection to highlight the more important ideas and concepts.

Appendix 5.A Multimedia Material

The contents of the videos referenced through the book are detailed in this appendix, enumerated in the following list. All videos can be found on the Internet, at the multimedia page of the author’s website: http://www.angelsantamaria.eu.

• 1. High-frequency MAV state estimation using inertial and optical flow measurement units. This video accompanies Chap. 2 and demonstrates the odometry estimation method through real MAV flights. First, experiments benchmarking all filter types against ground-truth (i.e., a motion capture system) are presented. Second, the video shows the technique with flights in a GPS-denied outdoor scenario. All flights are done using the hardware setting A described in Sect. 2.10.2. This video is related to [7].

• 2. Autonomous navigation of micro aerial vehicles using high-rate and low-cost sensors. In contrast to Video 1, Video 2 shows results of using the odometry estimation technique described in Chap. 2 using setting B —see 2.10.2). This video supports [6]. In this case, all algorithms are running on-line and on board the MAV, and the estimated state is used to feed the nonlinear controller presented in Sect. 2.9.

• 3. Uncalibrated image-based visual servoing. This video supports the uncalibrated visual servo method described in Sect. 3.6 and accompanies the paper [3]. The technique outperforms calibrated visual servo schemes in situations with noisy calibration parameters and for unexpected changes in the camera zoom. The method’s performance is demonstrated both in simulations and in a ROS implementation of a quadrotor servoing task.

• 4. Hybrid visual servoing with hierarchical task composition for aerial manipulation. The video shows real experiments of the hybrid visual servoing technique presented in Sect. 4.7.1. A hierarchical control law is used to position the UAM end effector with a PBVS while keeping the target in the camera FoV with an IBVS. Moreover, two subtasks are active: A task to vertically align the arm CoG with the platform gravitational vector, and an other task to reach desired arm joint positions. The video presents four main maneuvers: two grasping and two plugging operations with bars. This video accompanies the paper [1].

• 5. Task priority control for aerial manipulation. The hierarchical task controller described in Sect. 4.4.2 is validated in this video through simulation case studies as in the experiments Sect. 4.7.2.1. The mission shown consists on servoing the UAM end effector using visual information for an inspection and surveillance task. The related publication is [5].

• 6. Uncalibrated visual servo for unmanned aerial manipulation. This video shows experiments similar to those presented in Video 5, but this time with a real UAM —see experiments Sect. 4.7.2. The effect of adding hierarchically subtasks (i.e., using the control law presented in Sect. 4.4.2) is specifically shown. This material is related with [4].

• 7. Trajectory generation for unmanned aerial manipulators through quadratic programming. In Sect. 4.5 we presented a technique which applies quadratic programming (i.e., the on-line active set strategy) to generate a feasible joint trajectory, in order to accomplish a set of tasks with defined bounds and inequality constraint. This video accompanies the Sect. 4.5 and shows a mission composed of two phases, navigation and interaction, performed by an aerial manipulator. Weights of the cost functions are assigned in order to perform the two phases with different strategies. During the navigation phase, arm joints are guided towards a desired configuration for the sake of stability and motion is performed mainly with quadrotor DoFs. On the other hand, during the interaction phase, motion is performed by robot arm joints to obtain more accuracy. This video is related with [2].