An Adaptive Imitation Learning Framework for Robotic Complex Contact-Rich Insertion Tasks

Complex contact-rich insertion is a ubiquitous robotic manipulation skill and usually involves nonlinear and low-clearance insertion trajectories as well as varying force requirements. A hybrid trajectory and force learning framework can be utilized to generate high-quality trajectories by imitation learning and find suitable force control policies efficiently by reinforcement learning. However, with the mentioned approach, many human demonstrations are necessary to learn several tasks even when those tasks require topologically similar trajectories. Therefore, to reduce human repetitive teaching efforts for new tasks, we present an adaptive imitation framework for robot manipulation. The main contribution of this work is the development of a framework that introduces dynamic movement primitives into a hybrid trajectory and force learning framework to learn a specific class of complex contact-rich insertion tasks based on the trajectory profile of a single task instance belonging to the task class. Through experimental evaluations, we validate that the proposed framework is sample efficient, safer, and generalizes better at learning complex contact-rich insertion tasks on both simulation environments and on real hardware.

Modular Piezoresistive Smart Textile for State Estimation of Cloths

Smart textiles have found numerous applications ranging from health monitoring to smart homes. Their main allure is their flexibility, which allows for seamless integration of sensing in everyday objects like clothing. The application domain also includes robotics; smart textiles have been used to improve human-robot interaction, to solve the problem of state estimation of soft robots, and for state estimation to enable learning of robotic manipulation of textiles. The latter application provides an alternative to computationally expensive vision-based pipelines and we believe it is the key to accelerate robotic learning of textile manipulation. Current smart textiles, however, maintain wired connections to external units, which impedes robotic manipulation, and lack modularity to facilitate state estimation of large cloths. In this work, we propose an open-source, fully wireless, highly flexible, light, and modular version of a piezoresistive smart textile. Its output stability was experimentally quantified and determined to be sufficient for classification tasks. Its functionality as a state sensor for larger cloths was also verified in a classification task where two of the smart textiles were sewn onto a piece of clothing of which three states are defined. The modular smart textile system was able to recognize these states with average per-class F1-scores ranging from 85.7 to 94.6% with a basic linear classifier.

Tunable Neural Encoding of a Symbolic Robotic Manipulation Algorithm

We present a neurocomputational controller for robotic manipulation based on the recently developed “neural virtual machine” (NVM). The NVM is a purely neural recurrent architecture that emulates a Turing-complete, purely symbolic virtual machine. We program the NVM with a symbolic algorithm that solves blocks-world restacking problems, and execute it in a robotic simulation environment. Our results show that the NVM-based controller can faithfully replicate the execution traces and performance levels of a traditional non-neural program executing the same restacking procedure. Moreover, after programming the NVM, the neurocomputational encodings of symbolic block stacking knowledge can be fine-tuned to further improve performance, by applying reinforcement learning to the underlying neural architecture.

Deep Reinforcement Learning-Based Robotic Grasping in Clutter and Occlusion

In robotic manipulation, object grasping is a basic yet challenging task. Dexterous grasping necessitates intelligent visual observation of the target objects by emphasizing the importance of spatial equivariance to learn the grasping policy. In this paper, two significant challenges associated with robotic grasping in both clutter and occlusion scenarios are addressed. The first challenge is the coordination of push and grasp actions, in which the robot may occasionally fail to disrupt the arrangement of the objects in a well-ordered object scenario. On the other hand, when employed in a randomly cluttered object scenario, the pushing behavior may be less efficient, as many objects are more likely to be pushed out of the workspace. The second challenge is the avoidance of occlusion that occurs when the camera itself is entirely or partially occluded during a grasping action. This paper proposes a multi-view change observation-based approach (MV-COBA) to overcome these two problems. The proposed approach is divided into two parts: 1) using multiple cameras to set up multiple views to address the occlusion issue; and 2) using visual change observation on the basis of the pixel depth difference to address the challenge of coordinating push and grasp actions. According to experimental simulation findings, the proposed approach achieved an average grasp success rate of 83.6%, 86.3%, and 97.8% in the cluttered, well-ordered object, and occlusion scenarios, respectively.