Closed Loop Robotic Arm Control for Agricultural Applications

There are several barriers preventing mechanization of tasks that require detailed interactions, such as pruning and picking, with specialty crops. While each application requires unique solutions, this project addresses the critical system level issue of how to move a robot arm around and through complex plants and perform specific tasks in a dynamic environment. Real plants are all unique with constantly changing branches, foliage and fruit. The economic viability of mechanization requires very fast action, but it is difficult to delicately maneuver a robot arm (mounted on a platform moving down the row) through such complex plants. For several years, Vision Robotics has been developing systems that use stereo cameras to model plants. Using this technology, robots can now "understand" the plants well enough to perform the various tasks. However, existing commercial and research systems that control arm motions are either significantly too slow or expensive for economically viable systems. The objective of this project is to overcome these deficiencies by building on Phase 1 results by researching and developing the key systems to improve the control of arm motions. Dynamically reacting to changes and errors requires closed-loop control of the robotic arm, which has been successfully implemented for many non-agricultural applications (e.g., spot-welding) and even some agricultural applications (e.g., hedge-pruning) that only require coarse and/or slow control of an actuator. Unfortunately, existing closed-loop systems are not suitable for high-speed and detailed operations on complex objects such as those required for pruning and picking. Research and development efforts in Phase II will focus on building on an existing stereo vision based arm controller currently deployed on the company's grapevine pruner prototype by developing robust, computationally fast and accurate methods of calculating the difference between the current and desired position and orientation of robot arms and using that feedback to control commercially available off-the-shelf robot arms. Coupling Vision Robotics' existing plant modeling technology with the high speed and accurate, yet low cost, arm control systems developed during this project will create a viable system for working on specialty crops. In particular, the grapevine pruner upon which the control system will be demonstrated, will ultimately prune as well as manual labor at an equal or lower cost. Currently, manual pruning costs US growers approximately $700M per year and the availability of manual labor presents a potential crisis. Once fully implemented, this intelligent mechanization gives growers a third option when deciding whether to use a shrinking and more expensive labor pool or lower quality mechanization.