Achieving full autonomy in aerial and physical human-robot interaction control via onboard perception algorithms relying on computer vision

Finished: 2022-06-16

MSc assignment

In the context of aerial physical interaction, a branch of aerial robotics focused on the study and the accomplishment of contact-based interaction and manipulation tasks with aerial robots, the goal of this MSc thesis assignment is to enhance the level of autonomy of the aerial robots developed at RAM-UT, which currently rely mainly on Motion-Capture (MoCap) state feedback for indoor localization. After an initial extensive analysis of existing Visual Inertial Odometry (VIO) and Visual Simultaneous Localization and Mapping (V-SLAM) techniques in the literature, the candidate should select the most promising implementations, compare them, and choose the one that fits best with the given requirements. During the implementation phase, the candidate should integrate the chosen software component in the overall software architecture used in the RAM-UT group and extensively test it in real experiments.

The research questions starting from which the assignment should be developed can be (others can be investigated):
  • which are the hardware (sensors) and software components that allow an aerial robot to autonomously perform an object handover (possibly, to a human)?
  • which are currently the most promising localization software component to perform onboard localization and navigation?
  • is the precision in the localization solution (indoor and outdoor) relying only on cameras enough to perform an autonomous aerial handover to humans?

The final demonstration of the thesis should showcase the capabilities of the flying manipulator (consisting of an aerial platform endowed with an arm manipulator) in localizing and navigating in the surrounding environment (indoor and outdoor) providing and getting objects from/to human workers using onboard cameras as main sensorial feedback.
This is supported by a large European Project called Aerial-Core, in which the RAM-UT research group is a Work Package leader, cf. [1]. The main goal of Aerial-Core is to develop an integrated aerial cognitive robotic system with unprecedented capabilities on the operational range and safety in the interaction with human co-workers for applications such as the inspection and maintenance of large linear infrastructures.

The main goals of this MSc thesis assignment are:
1) studying and finding which is the best vision-based localization/navigation algorithm available currently in the literature and integrate it into the RAM-UT aerial manipulation software framework (which is based on an extension of ROS, called GenoM, cf. [2])
2) studying and finding the best way to recognize the human in the scene, its main body, and its hand, and localize the robot with respect to it, using the onboard camera that we have on the RAM-UT aerial robot (a software package in ROS is already available from some partners in the Aerial-Core project, it should be tested and integrated).
3) studying and finding the best way to put together these perception algorithms with the RAM-UT control framework (developed by other people in the team) that is able to control the aerial manipulator and its end-effectors so that vision-based aerial robotic handover and object pickup can become reality (the final demonstration for this thesis can be simplified and should be achieved with incremental steps).

The assignment has also important programming and integration aspects, in addition to the aforementioned scientific aspect. In fact, in order to obtain a successful master thesis, the student will need, among other things, to program in C++, use ROS, learn the meta-middleware extension of ROS used at RAM-UT (GenoM), and be familiar with computer vision algorithms and libraries. The student will need to use simulators such as Gazebo and real hardware like already available cameras (intel Realsense T265 and D435i - others can be envisioned).

References:
[1] https://aerial-core.eu/ && https://www.ram.eemcs.utwente.nl/research/projects/aerial-core
[2] https://git.openrobots.org/projects/genom3
[3] http://rpg.ifi.uzh.ch/docs/Encyclopedia19VIO_Scaramuzza.pdf
[4] https://ieeexplore.ieee.org/document/8460664