• No results found

• Housing the product database of the shop including the products’ relative position in the topological area they are contained in

• Supporting the shopping process by managing shopping lists, recipes, and so forth

More information on these modalities can be found in the corresponding section in the HRI-related chap- ter (Chap. 6.2). The communication layer communicates with the strategic layer using the command

interface as described in Chapter 3.3.3 “The command interface”.

As the alternative implementations of the communication layer are not part of the core part of the control system, they are introduced in the Appendix. Two implementations have used with InBOT: For this thesis the InBOT-UI (see Appendix C.4) was developed, implementing speech output, a bar code scanner, as well as a shopping list and recipe management via touch screen. In addition, a second communication layer was integrated, namely the CR-UI (see [87] and Appendix C.5), developed at TU Vienna.

4.8. Experiments and evaluation

Following the description of the individual components of the navigation system (including the individual components’ evaluation), this section presents three tests which have been conducted using the complete system – the last two tests even involving users recruited “from the street”. The focus of this section will be on the test results regarding the navigation system. In the chapter focusing on HRI (Chap. 6) the tests will be picked up again with an human-factors point of view.

4.8.1. First system test

One example out of the tests performed with the complete system is presented here, supplementing the results provided in the in individual sections dedicated to the individual behaviors or functionalities, re- spectively. For the sake of reproducability, this test has been performed in a simulated environment – the real world tests involving users are described in the subsequent sections. The simulation of the robot plat- form is very detailed and accurate down to each individual scan point of the laser scanners. Additionally, the simulation – for which large pieces could be gratefully taken from previous work – incorporates the (de-) acceleration capabilities, the drive system and the S300 laser scanners including the corresponding precisions and variances. The simulation does not interfere with the remaining control system, as it just substitutes the Hardware Abstraction Layer, using exactly the same interface.

Figure 4.50 shows this test: The robot was given the tasks to visit a list of products and was impaired by lots of obstacles. All components described in this chapter have been used, from the communication

layer down to the safety reflexes. The robot passed dead-ends, narrow passages, and evaded large and tiny

obstacles. The bottom part of the figure shows the robot’s knowledge of the experiment: the topologic-

metrical map, the local sensor readings, and the path the robot has taken.

4.8.2. Second system test

This first large-scale test performed with untrained users was a collaborative activity with KTH Stockholm and TU Vienna. It aimed at gathering first impressions on the HRI system in general and the shopping list assistant (commanded by the CR-UI) in particular, thus, the setup was reduced to short linear runs and the

Fig. 4.50.: This figure shows a test run performed with the robot using the MCA simulation environment. The robot was given the task to visit a list of products (located at the places marked with the numbers 21, 19, 11, 2, and 5). The top figure shows the (simulated) real world including lots of obstacles. The bottom figure shows the robot’s knowledge. According to the defined requirements, the robot’s knowledge is limited to the current sensor readings (green dots) and the topologic-metrical map which is indicated by the blue lines (borders of topological areas) and the green rectangles (locations of the RFID barriers). Even tough relying only on this few information, the robot successfully reached all products and evaded a large number of obstacles, including a dead-end.

available functionalities were reduced to a basic set. The results are briefly summarized in the Figure 4.51 (top).

4.8.3. Third system test

This third test was the second test with users – again a collaborative action with KTH and TUW. It was performed with full complexity and functionality (omitting moving obstacles). In particular, this time the geometrical scene analysis has been used.

The test was performed with 2+10 untrained users. Two pilot runs have been performed a priory to adapt parameters of the navigation system to the environment. Otherwise, the system was identically in all twelve runs. Initially, the users had to enter a scripted shopping list containing 12 products using the touch screen user interface implemented by the CR-UI. Then they had to order the robot to guide them to the first nine products. The last three products should be accessed by driving the robot in the Manual Steering

4.8. Experiments and evaluation

Fig. 4.51.: Facts on the second and third system test on InBOT conducted with users.

Mode. While maneuvering in a narrow corridor the user had to avoid a box which was dropped in his path

in cooperation with the obstacle avoidance assistant. A sketch of this setup is provided together with a summary of the results in Figure 4.51 (bottom).

From a navigation and control system point of view, the test can be considered successful: all users were able to finish all tasks and we observed no collisions at all and hardly any wrong movements. The success rate of the navigation tasks was 97.2% with only 3 errors: one time the robot moved in an incorrect direction. Most probably the corridor InBOT should drive into seemed to be blocked (by the user and a person from the experimental staff steering a cart with a camera mounted on top). Hence, the predictive obstacle avoidance tried to find another way. But this could not be verified without stopping the experiment, thus, it counts as

mistake. In the two other cases the robot did not start moving even though the speech output “I will guide you to ...” was uttered. In all three cases the user could solve the problem by the canceling and repeating the command. On the second try the robot executed the task correctly.