Dynamic On-Palm Manipulation via Controlled Sliding

TL;DR

This work addresses fast, dynamic non-prehensile manipulation by enabling controlled sliding through a contact-implicit MPC (C3) framework. A Linear Complementarity System captures end-effector, tray, and external contacts, while a robot-only model handles low-level tracking via an Operational Space Controller; the method leverages ADMM-based complementarity handling to plan stick-slip transitions in 3D without reference trajectories. Key contributions include integrating C3 with a downstream OSC for dynamic tasks, validating on a 3D tray retrieval/lift/place scenario, and demonstrating robustness to modest model inaccuracies and external perturbations. The approach achieves fast, multi-contact manipulation in under 5 seconds and generalizes to tasks such as rotating a tray with an external wall, highlighting its practical impact for dynamic, contact-rich manipulation in robotics.

Abstract

Non-prehensile manipulation enables fast interactions with objects by circumventing the need to grasp and ungrasp as well as handling objects that cannot be grasped through force closure. Current approaches to non-prehensile manipulation focus on static contacts, avoiding the underactuation that comes with sliding. However, the ability to control sliding contact, essentially removing the no-slip constraint, opens up new possibilities in dynamic manipulation. In this paper, we explore a challenging dynamic non-prehensile manipulation task that requires the consideration of the full spectrum of hybrid contact modes. We leverage recent methods in contact-implicit MPC to handle the multi-modal planning aspect of the task. We demonstrate, with careful consideration of integration between the simple model used for MPC and the low-level tracking controller, how contact-implicit MPC can be adapted to dynamic tasks. Surprisingly, despite the known inaccuracies of frictional rigid contact models, our method is able to react to these inaccuracies while still quickly performing the task. Moreover, we do not use common aids such as reference trajectories or motion primitives, highlighting the generality of our approach. To the best of our knowledge, this is the first application of contact-implicit MPC to a dynamic manipulation task in three dimensions.

Figures (11)

• Figure 1: We examine a dynamic sliding task, where the robot uses the full spectrum of contact modes (sticking, sliding, making and breaking contact) in order to retrieve a tray resting on external supports. We use contact-implicit MPC to automatically plan when and where to use different contact modes. With careful consideration on how to integrate the simplified MPC model with the robot arm, we are able complete the entire maneuver of retrieving the tray, lifting it, and placing it back on the external supports in just 5 seconds, demonstrating dynamic capability for a contact-rich task.
• Figure 2: The three target positions. The grasp locations on the tray change between targets, thus requiring the end effector to either slide and/or break contact with the tray.
• Figure 3: We abstract the system into two models. The LCS model captures the contact forces $\lambda$ between the end effector, tray, and supports. In the LCS model, the robot arm is abstracted away and replaced with direct inputs to the end effector $u_{lcs}$. We then use a robot-only model to track the end effector position $q_{ee}(t)$ and force $u_{lcs}(t)$ trajectories commanded from the MPC, so $\lambda_{ee} = u_{lcs}$.
• Figure 4: We consider seven total contacts for our task. The contact geometries shown in red. We represent the tray as a cylinder and we choose fixed contact points on the end effector and supports, which we model as spheres. The radii for the spheres are enlarged by a factor of 10 for visibility purposes. A minimum of three contact points are required to approximate surface-surface contact between the end effector and tray, while two contact points are required to model each line contact from the supports.
• Figure 5: System diagram for the hardware implementation. The different colored boxes indicate separate processes which are connected via arrows that indicate represent communication via ROS/LCM.