Blended Shared Control of a Hydraulic Excavator

The situational awareness, robustness and expertise unique to skilled operators have precluded the full automation of many construction tasks. However, automatic control may be able to execute some quasi-repetitive construction tasks more efficiently than a human. This dichotomy has led to interest in Shared Control (SC) which synthesizes the advantages of both autonomous and manual control of machinery. A particularly interesting case of SC for unstructured environments is called Blended Shared Control (BSC) where weighted inputs from the human operator and autonomous controller are continuously combined to perform a task. We discuss the kinematics and dynamics of an excavator, and some low-level control and coupling issues. Then, we present a 1/12th scale hydraulic excavator test-bed, which is utilized for shared control experimentation. We introduce a state-machine based prediction framework and a BSC implementation which can be executed in real-time on embedded hardware. The proposed framework uses subgoals, distinct points in the workspace which the operator must reach in order to complete the task. We then analyze the performance of this BSC implementation for a typical earth-moving task with several operators on the excavator test-bed. Digging efficiency of each operator increases under BSC, whereas the effect of BSC on cycle time varies with operator skill level and operation style.