Validation of Subsystem Functionalities

During the field tests in Utah two major robot functionalities that are needed for a sample return mission have been tested: payload pickup and soil sampling. The payload pickup pro- cedure has been developed by Sankaranarayanan Natarajan in cooperation with the author of this thesis based on the works by Roehr, Cordes, and F. Kirchner (2014), and the soil sampling procedure has been developed by Sankaranarayanan Natarajan and tested and incrementally improved with the author of this thesis (cf. (Brinkmann, Cordes, et al.2018; Brinkmann, Roehr, et al.2018)).

Payload Pickup

Payload pickup is an essential element for reconfiguration of the multi-robot team and it relies on a visual servoing approach. Figure 6.10shows the general image data from the manipu- lator’s camera during the process of a payload pickup. The payload pickup sequence failed under the presented conditions due to hard shadows of theEMIwhich prevented the detection of the inner two markers. The inner marker set is required for the final docking of the end effector to the payload due to the limited field of view of the camera in theEMI, and the field of view additionally narrows when the end effector approaches theEMI.

This outdoor experiment illustrated the lighting condition of a target environment and showed a point a failure as the result of augmenting theEMIwith an additional marker set. Lighting conditions had been initially validated, although only for the initial (outer) marker set as illus- trated in Figure6.11. The following, intermediate addition of two markers intended to increase

(a)Data exchange between robots (local control centre) and mission control.

(b)Direct data exchange between robots.

Figure 6.9: Communication volumes for inter-site and inter-robot communication. The com- parison between letter and actual message content size for the inter-robot communication shows the minimal protocol overhead in a real application.

(a)Start approach by identification of the outer marker set.

(b) End effector approaches and only one inner marker remains de- tectable.

(c) Shadow of EMI guidance pins prevents marker detection.

(d)Corresponding binary image af- ter thresholding.

Figure 6.10:Camera images augmented with the marker detection result during the approach of the end effector to pickup a payload item. Images (a) and (b) are taken during the same approach. Images (c) and (d) refer to an approach at later time of day, indicated by the shadows of the interface pins.

(a) The initial marker setup with only an outer marker set during gener- ation of test data under extreme lighting condi- tions.

(b) The corresponding bi- nary image after thresh- olding.

(c) Final marker setup us- ing a highly redundant set of markers.

Figure 6.11:Initial and final marker setup for theEMI.

the precision of the overall approach, but at the same time introduced additional restrictions and effectively a point of failure. To increase the robustness of the visual servoing process the inner marker setup has been adapted as illustrated in Figure 6.11c. This final marker setup comes with an increased marker redundancy and thereby improves the detection of markers, so that the general 90◦ rotational invariance of theEMIcan be maintained; the setup with two markers has limited this rotational invariance. An additional benefit of the redundant marker setup is also an improved precision of the detected pose. Figure6.11a indicates the detected pose by a coordinate system draw into the camera image. The origin of this coordinate system lies ideally in the base of the central pin of theEMI, yet it can be seen that the detection of a single marker can be insufficient for a precise detection of this pose. Increasing the number of detected markers is therefore necessary to achieve a sufficient precision of the visual servoing approach. Section6.2presents the success rate of the adapted approach.

Soil Sampling

Soil sampling is an example of an actively designed superadditive functionality. Several spe- cialised payload items have been implemented by Brinkmann, Cordes, et al. (2018) to illustrate the extensibility of reconfigurable multi-robot systems. The sampling functionality requires the mobile manipulation agent SherpaTT and a payload item which has been equipped with a soil sampling system. The so-called sampling module can be attached to SherpaTT’s end effector and a procedure control the sampling process. Firstly, the sampling module is low- ered to the ground until ground contact is established. Ground contact is identified through a force torque sensor in the manipulator, i.e., a certain threshold of force in z-direction has to be met. The identification of ground contact triggers a dragging of the sampling module towards its body centre in order to fill the shovel. After a predefined distance of dragging the shovel system is closed and the sampling module is lifted. Figure6.12shows the test setup in Utah. The implementation of the soil sampling application which has been tested in Utah showed that the sampling module has a too limited opening angle: the manipulator had to be tilted

(a)Sampling module is moved towards the ground. (b) Sampling module is dragged over the ground.

Figure 6.12:Soil sampling in Utah using a dedicated soil sampling module.

slightly upwards in order to compensate for this limited opening angle in order to successfully sample soil. In effect, the sampling module could not be used for sampling in flat terrain and the design had to be revised. The revised version allows the manipulator’s last link to remain parallel to the ground.