Seite - 204 - in Proceedings - OAGM & ARW Joint Workshop 2016 on "Computer Vision and Robotics“

actions and his/her activities, robot’s planned trajectories and positions of the robot’s tool will be considered as task relevant features to model the states of the HMM [2]. 2.3. Trajectory Planner The trajectory planner considers the static workcell, actual skeletal joint positions, detected human activities and 3D locations of the objects of interest as an input and computes a collision-less trajectory for the robot. These activity dependent collisions-less trajectories will result in process- specific object manipulations like Grasp, Lift, Place, and Present. During the execution of the process, trajectory planner will send status updates about the object manipulations which will be requested by the high-level planner. Path planning algorithms which were applied in [11], [7] will be investigated to verify which one of them would be ideal for safe execution of the considered process. 2.4. High-Level Planner High-Level Planner is an intermediate layer which receives the status updates continuously from major building blocks and robot’s tool positions to monitor the process execution. The High-Level Planner will be included with the static description of sequential order of process steps/tasks involved within a process. During the execution of the process, the High-Level Planner will compare the actual state of the process with its desired state and identify the low-level process deviations from the robot assistant side. Based on these deviations, the trajectory planner will then compute a collision-free trajectory which will lead to successful completion of the previously failed process steps/tasks. Here, High-Level Planner will continuously send the same process step/task to the trajectory planner until the identified deviation vanishes. 3. Experimental Setup The process of assembling a Steam cooker device using its individual objects is considered here. The individual objects of the steam cooker are present on the worktable as depicted below. Figure 3: Experimental setup included with individual objects of a steam cooker device In Figure 3, UR10 is the universal robot which is placed on a movable platform. This movable platform is clamped to the worktable where the human worker and ur10 robot will share the workspace. A Kinect v2 sensor is applied for the human action and human activity recognition and 204