Grasping Moving Objects: Adaptive Motion Through Tactile Sensing

Abstract

The aim of this project is to explore how adaptive grasping motions, informed by real-time tactile feedback, can improve grasp robustness when grasping moving objects. It was hypothesised that an adaptive grasping motion, based on real-time tactile sensor feedback, could mitigate the effects of errors in the interception of a ball by a robotic gripper, resulting in a more robust control strategy for grasping moving objects.

An experimental methodology was developed to enable a systematic examination of a gripper's ability to grasp a moving object, under a range of grasping conditions. A virtual environment was developed which implemented this methodology and evaluated the performance of two grasping control strategies, a traditional, predictive strategy and heuristic-based, reactive strategy. A physical experimental apparatus was also created, and the results of simulated testing where verified in the real-world. Finally, an additional reactive control strategy, which relied on a neural network agent trained in simulation, was developed. The agent was deployed on the real-world gripper and its performance quantified in accordance with the same methods used in previous experiments.

Testing in simulation revealed that the reactive grasping strategy outperformed its predictive counterpart, achieving an average grasping success rate of 58% compared to 36% across the same range of grasping conditions. These results are echoed during testing with the physical apparatus. The predictive strategy had a grasp success rate of 71% while the heuristic based reactive strategy achieved 80%. The best robustness was attained by the neural network based strategy at 90% grasp success. These results demonstrate the additional grasp robustness achieved by an adaptive grasping motion, informed by real-time sensor data.

This thesis presents foundational research regarding this type of novel, adaptive grasping strategy. The strategies proposed demonstrate a significant improvement to a robot s ability to grasp moving objects in unstructured environments and enables exciting new applications in this field.