Compliant Grasping for Unstructured Environments

Aaron M. Dollar, Robert D.Howe

Support provided by the Office of Naval Research

Project Overview
One of the central challenges of robotics is grasping and manipulating objects in unstructured environments, where object properties are not known a priori and sensing is prone to error. The resulting uncertainty in the relationship between the object and gripper makes it difficult to control contact forces and establish a successful grasp or accurately position the object. One approach to dealing with this uncertainty is through compliance, so that positioning errors do not result in large forces and the grasper conforms to the object. This has most often been implemented through active control of manipulator impedance, and many studies have been devoted to impedance analysis and control techniques for robot arms and hands. This approach is based on active use of joint sensors for position, velocity and force/torque. An alternative approach is the use of mechanical compliance in the manipulator structure. Ideally, carefully designed passive compliance can simplify the grasping process by eliminating the need for a good deal of traditional sensor-based control.

Simulation to Optimize Grasper Configuration
In contrast to manipulators for unstructured environments that rely on active control for compliance, we are interested in passive joint compliance that results in large joint deflections and low contact forces, thus minimizing disturbance or damage to objects during the first phases of acquisition. In particular, we examine the performance of a simple gripper with two fingers, each with two revolute degrees of freedom (see figure). This gripper is perhaps the simplest configuration that is able to grasp a wide range of objects. We assume that the links are rigid lines between joints and that each joint of the gripper includes a passive linear spring in series with an actuator. Our goal is then to determine how variations in the joint stiffnesses and initial rest angles affect the ability to grasp objects. This simple configuration allows detailed analysis of parametric trade-offs, which is difficult for complex anthropomorphic hands. Performance is compared on the basis of the maximum range of object size and location that can be successfully grasped and the magnitude of contact forces. The results are analyzed to determine the ways that compliance and kinematic configuration contribute to grasping performance without the need for extensive sensing. Details of this work can be found here.


Experimental Validation
In order to experimentally validate the results of the simulation, we built a prototype grasper with the same kinematics as the simulated mechanism. The grasper is reconfigurable to allow investigation into how variations of rest angles and joint stiffness affect the performance. Optical encoders allow measurement of joint angle and testing for object enclosure and interchangeable metal torsional springs are mounted in each joint to provide passive compliance. Results from experimental work with this grasper corroborate the results from the simulation study. Details of this work can be found here.

Joint Coupling Design of Underactuated Grippers
We also have examined the nature of joint coupling in underactuated grippers for environments where object size and location may not be well known. The grasper considered in previous studies (above) was simulated as it was actuated after contact with a target object. The joint coupling configuration of the gripper was varied in order to maximize successful grasp range and minimize contact forces for a wide range of target object size and position. A normal distribution of object position was assumed in order to model sensing uncertainty and then weight the results accordingly. Proximal-distal joint torque ratios of around 0.6 produced near-optimal results for cases in which sensory information available for the task was poor, and ratios of around 1.0 produced good results for cases in which sensory information available for the task was good. Details of this work can be found here.

Future Work
Future work on this project includes work on evaluating tradeoffs between sensory suites of varying complexity, and more extensive experimentation with grasper prototypes.

Primary Publications

For more information contact Aaron Dollar,

Harvard BioRobotics Laboratory Home