Goal directed multi-finger manipulation: Control policies and analysis

Abstract

We present a method for one-handed, task-based manipulation of objects. Our approach uses a mid-level, multi-phase approach to organize the problem into three phases. This provides an appropriate control strategy for each phase and results in cyclic finger motions that, together, accomplish the task. The exact trajectory of the object is never specified since the goal is defined by the final orientation and position of the object. All motion is physically based and guided by a control policy that is learned through a series of offline simulations. We also discuss practical considerations for our learning method. Variations in the synthesized motions are possible by tuning a scalarized multi-objective optimization. We demonstrate our method with two manipulation tasks, discussing the performance and limitations. Additionally, we provide an analysis of the robustness of the low-level controllers used by our framework.

Relaxed-Rigidity Constraints: In-Grasp Manipulation using Purely Kinematic Trajectory Optimization

“…Andrews and Kry [1] take a hierarchical approach to inhand manipulation by splitting the problem into three phases: approach, actuation, and release. Their method uses an evolutionary algorithm to optimize the individual motions.…”

“…Research in inhand manipulation has focused largely on using full knowledge of the mechanical properties of the objects of interest in finding solutions [1,12,22,23]. This reliance on object specific modeling makes in-hand manipulation expensive and sometimes infeasible in real-world scenarios, where robots may lack high-fidelity object models.…”

Towards learning hierarchical skills for multi-phase manipulation tasks

“…Previous work in robotics has already shown the benefits of incorporating phases into the design of controllers [32], [10], and several methods have been proposed for learning controllers for multi-phase tasks [22], [26], [1], [28]. Levine el al.…”

Learning to predict phases of manipulation tasks as hidden states

“…Previous work on using phases of manipulation tasks has largely assumed that the phases are predefined [7], [15], [1]. Debus et al estimate the contact state for a peg-inhole task using an HMM, but they assume the network of contact states and the descriptions of the states are given [7].…”

“…The transitions between these phases occur after the robot detects specific tactile events. Andrews and Kry proposed a controller for performing in-hand manipulation based on a three-phase structure [1].…”