We present a comprehensive approach to handle perception uncertainty to reduce failure rates in robotic bin-picking. Our focus is on mixed-bins. We identify the main failure modes at various stages of the bin-picking task and present methods to recover from them. If uncertainty in part detection leads to perception failure, then human intervention is invoked. Our approach estimates the confidence in the part match provided by an automated perception system, which is used to detect perception failures. Human intervention is also invoked if uncertainty in estimated part location and orientation leads to a singulation planning failure. We have developed a user interface that enables remote human interventions when necessary. Finally, if uncertainty in part posture in the gripper leads to failure in placing the part with the desired accuracy, sensor-less fine-positioning moves are used to correct the final placement errors. We have developed a fine-positioning planner with a suite of fine-motion strategies that offer different tradeoffs between completion time and postural accuracy at the destination. We report our observations from system characterization experiments with a dual-armed Baxter robot, equipped with a Ensenso three-dimensional camera, to perform bin-picking on mixed-bins.
Robotic bin picking requires using a perception system to estimate the posture of parts in the bin. The selected singulation plan should be robust with respect to perception uncertainties. If the estimated posture is significantly different from the actual posture, then the singulation plan may fail during execution. In such cases, the singulation process will need to be repeated. We are interested in selecting singulation plans that minimize the expected task completion time. In order to estimate the expected task completion time for a proposed singulation plan, we need to estimate the probability of success and the plan execution time. Robotic bin picking needs to be done in real-time. Therefore candidate singulation plans need to be generated and evaluated in real-time. This paper presents an approach for utilizing computationally efficient simulations for on-line evaluation of singulation plans. Results from physical experiments match well with predictions obtained from simulations.
Robotic bin picking requires using a perception system to estimate the pose of the parts in the bin. The selected singulation plan should be robust with respect to perception uncertainties. If the estimated posture is significantly different from the actual posture, then the singulation plan may fail during execution. In such cases, the singulation process will need to be repeated. We are interested in selecting singulation plans that minimize the expected task completion time. In order to estimate the expected task completion time for a proposed singulation plan, we need to estimate the probability of success and the plan execution time. Robotic bin picking needs to be done in real-time. Therefore, candidate singulation plans need to be generated and evaluated in real-time. This paper presents an approach for utilizing computationally efficient simulations for generating singulation plans. Results from physical experiments match well with predictions obtained from simulations.
In this paper, we address the complexities arising due to occlusions in robotic bin-picking. Our focus is on mixed-bins. Most traditional planners try to find collision-free paths to extract objects, while returning a failure whenever a collision is anticipated between the object to be extracted and a neighboring object occluding the former. We take a different approach in this paper. Our approach is inspired by the fact that simple motions of the grasped object may result in the transition of the object from a collision-state to a collision-free state. Our approach exploits the local geometric relationships between the objects in contact with each other and the change in these relationships as the grasped object is moved to make conservative predictions whether such motion results in tangle-free extraction. We demonstrate our approach using experiments with a Kuka LBR iiwa robot singulating objects from a pile of convex and concave objects.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.