Traditional robotic manipulation mostly focuses on collision-free tasks. In practice, however, many manipulation tasks (e.g., occluded object grasping) require the robot to intentionally collide with the environment to reach a desired task configuration. By enabling compliant robot motions, collisions between the robot and the environment are allowed and can thus be exploited, but more physical uncertainties are introduced. To address collision-rich problems such as occluded object grasping while handling the involved uncertainties, we propose a collision-inclusive planning framework that can transition the robot to a desired task configuration via roughly modeled collisions absorbed by Cartesian impedance control. By strategically exploiting the environmental constraints and exploring inside a manipulation funnel formed by task repetitions, our framework can effectively reduce physical and perception uncertainties. With real-world evaluations on both single-arm and dual-arm setups, we show that our framework is able to efficiently address various realistic occluded grasping problems where a feasible grasp does not initially exist.

Affordances are central to robotic manipulation, where most tasks can be simplified to interactions with task-specific regions on objects. By focusing on these key regions, we can abstract away task-irrelevant information, simplifying the learning process, and enhancing generalisation. In this paper, we propose an affordance-centric policy-learning approach that centres and appropriately \textit{orients} a \textit{task frame} on these affordance regions allowing us to achieve both \textbf{intra-category invariance} -- where policies can generalise across different instances within the same object category -- and \textbf{spatial invariance} -- which enables consistent performance regardless of object placement in the environment. We propose a method to leverage existing generalist large vision models to extract and track these affordance frames, and demonstrate that our approach can learn manipulation tasks using behaviour cloning from as little as 10 demonstrations, with equivalent generalisation to an image-based policy trained on 305 demonstrations. We provide video demonstrations on our project site: https://affordance-policy.github.io.

This study investigates learning from demonstration (LfD) for contact-rich tasks. The procedure for choosing a task frame to express the learned signals for the motion and interaction wrench is often omitted or using expert insight. This article presents a procedure to derive the optimal task frame from motion and wrench data recorded during the demonstration. The procedure is based on two principles that are hypothesized to underpin the control configuration targeted by an expert, and assumes task frame origins and orientations that are fixed to either the world or the robot tool. It is rooted in screw theory, is entirely probabilistic and does not involve any hyperparameters. The procedure was validated by demonstrating several tasks, including surface following and manipulation of articulated objects, showing good agreement between the obtained and the assumed expert task frames. To validate the performance of the learned tasks by a UR10e robot, a constraint-based controller was designed based on the derived task frames and the learned data expressed therein. These experiments showed the effectiveness and versatility of the proposed approach. The task frame derivation approach fills a gap in the state of the art of LfD, bringing LfD for contact-rich tasks closer to practical application.