Hardware and decentralized-control algorithms have been developed during continued research on the sensors, the actuators, and the design and functional requirements for systems of multiple mobile robots cooperating in the performance of tightly coupled tasks — for example, grasping and lifting long objects on challenging terrain. [Different aspects of the hardware and algorithms were described in “Advances in Cooperative Transport by Two Mobile Robots” (NPO-30376) NASA Tech Briefs, Vol. 26, No. 8 (August 2002), page 60. Although this research is oriented toward developing robotic capabilities for exploration of Mars, these capabilities could also be utilized on Earth.

Figure 1. Two Mars Rovers cooperate in carrying a long payload.

Two robots used in this research have been rovers designed for prototype Mars missions: the Sample Return Rover and the Sample Return Rover 2000 (see Figure 1). Robotic arms (manipulators) on both rovers were modified to include grippers redesigned as their end effectors. Each gripper (see Figure 2) is of a simple yet novel design, incorporating three interlocking digits: Two fingers facing one side straddle a thumb that faces the opposite side, and these opposing digits are driven at their respective bases by counterrotating, parallel pivots. The finger and thumb pivots are driven through a transmission actuated by a single DC motor equipped with an incremental- shaft-angle encoder for feedback. The fingers and thumb are hooked at their tips to provide positive retention against slippage of a grasped payload. The finger geometry accommodates a variety of payloads of any general cross section narrow enough to fit between the fingers.

The design of the grippers provides passive compliance to simplify control of the gripping process: A spring loaded pivot at the base of each digit enables flexing to “open up the grip,” when the payload exerts an overload on the gripper. In the absence of a payload, mechanical hard stops and limit switches arrest the gripper and signal its closure. The compliant joint on each digit also includes a switch that provides limited feedback related to the gripping force, by signaling the point at which each digit flexes open. Another sensor that provides feedback is located in the “palm” of the gripper: a compliant bumper on a switch provides information related to the location of the payload; that is, whether or not the payload has been grasped.

The cooperative grasping and lifting behavior of the robots is characterized as being partitioned into the following four distinct phases:

Figure 2. A Gripper includes two fingers that straddle and oppose a thumb. The fingers and thumb are connected at their bases to compliant pivots counterrotated by a motor and transmission.
  1. Visual Target Search Phase

    In this phase, cameras on the rovers are used to search and capture images of unique patterns placed at specified positions on the payload. From the images, estimates of the Cartesian coordinates of the point of grasp and of an approach unit vector are computed for each rover. Estimates of the distance each rover must travel and the orientation that each rover must assume to place the payload within the work space of its manipulator are also computed. Synchronization occurs between the rovers, and both rovers proceed to position and orient themselves within their respective manipulator work spaces. The rovers then proceed to the approach phase.

  2. Approach Phase

    The gripper of each rover is moved along the approach unit vector toward the point of grasp computed during the visual target search phase, until the palm contact switch is triggered. If contact is made as indicated by the thumb or finger switch during the approach, the gripper is moved away from the payload until thumb or finger switch is reset and the approach phase aborted. This completes the approach phase. Synchronization occurs between the rovers before proceeding to the grasp phase.

  3. Grasp Phase

    The digits are moved in such a manner as to ensure that the finger and thumb contact switches are triggered, confirming a firm grip of the payload. The gripper is not equipped with force and torque sensors; instead, the spring preloads of compliant joints are adjusted manually ahead of time to set the approximate value of the gripping force or a prescribed amount of over travel is used to give an estimated gripping force. An “intelligent” heuristic, rule-based compliance-control algorithm is implemented by use of the limited feedback provided by the contact switches during grasping. If the thumb contact switch is closed independently of any finger contact switch, the arm is moved perpendicular to approach unit vector until the thumb contact switch is reset to the open condition. A similar maneuver is performed if either or both of the finger contact switches are triggered independently of the thumb contact switch. These compensatory maneuvers have been demonstrated to guarantee a firm grasp with rover positioning errors of as much as 5 cm.

  4. Payload Lift Phase

    In this phase, the payload is lifted about 20 cm in two stages: 5 cm in the first stage and 15 cm in the second stage. It is easy to develop control algorithms that use the information provided by the switches on the fingers and thumb, along with the information from the bump switch in the palm, to deduce whether the payload is within grasp; that help to center the gripper about the payload; and that deduce minimal information about the orientation of the payload. For example, by determining which finger is flexing, such an algorithm can help to determine the angle of the payload.

This work was done by Ashitey Trebi-Ollennu, Hari Das Nayar, and Anthony Ganino of Caltech for NASA’s Jet Propulsion Laboratory.