Page - 142 - in Proceedings - OAGM & ARW Joint Workshop 2016 on "Computer Vision and Robotics“

Figure5.25: This figure shows the inspection points in more detail. (a)Baxter inspecting the box with it’s endeffector camera. (b)Baxter inspecting the Box visualized in RViz. The white point cloud indicates the detected box. The orange voxels vi- sualize the collision scene. Figure5.26: This figure shows Baxter inspecting the box in reality as well as visualized in RViz. 72 (a) Baxter inspecting the box with it’s end-effector camera. Figure5.25: This figure shows the inspection points in more detail. (a)Baxter inspecting the box with it’s end ffector camera. (b)Baxter inspecting the Box visualized in RViz. The white point cloud indicates the detected box. The orange voxels vi- sualize the collision scene. Figure5.26: This figure shows Baxter inspecting the box in reality as well as visualized in RViz. 72 (b) Baxter inspecting the box visualized in RViz. White points indicate the detected box. Orange voxels represents the colli- sion scene. Figure5: Realizationof theprimitive ItemDetection. given setups. Details about the evaluations ca be found in [10]. In Figure 6 the results of the individual skill primitives are shown. The green bar indicates the successful execution rate, the gray bar marks the failure executions which are detected by the system and the red bar shows the failures whicharenotdetectedbythesystem. Evenif thesuccess rateofsomeprimitive isnotoverwhelming, the system detects the failure and reacts to it. The recognition of failures is one fundamental ability of reliable systems. Assoon as theerrors aredetected, the robot can react to it autonomously. de tec tH an dle gra sp Ha nd le mo ve To Su pp ort Po se pu llB ox mo ve Bo x de liv erO nT ray de tec tB ox de tec tIt em gra sp Ite m de liv erO nL ev el 0 20 40 60 80 100 Skillprimitive SkillprimitiveEvaluation Success DetectedFailure NotDetectedFailure Figure6: Skillprimitiveevaluation 6 142