High Performance Manipulation

by Christian Smith
picture

In order to study teleoperation in a highly dynamic context, we have constructed a fast robotic manipulator. This page will give a closer presentation of the technical specifications and implementation issues of the platform.

Table of contents

Design issues

The arm has been designed in order to use in experiments for teleoperated ball-catching. Therefore, the main goal of the mechanical design has been to cover a large enough area in short enough time to be able to do this.

A 'large enough area' has been defined as a 0.6 m by 0.6 m window, based on the precision of a normal underhand ball toss. A 'short enough time' has been defined as 0.5 s, based on the flight time of ballistic objects and the reaction times of the teleoperated system. In simulation our robot arm can, if started in the middle of the operating window, reach any point therein within 0.3 s, which should be fast enough with ample margin. In actual experiments, simulated performance has only been verified for a very limited amount of movements, but the actual system shows good promise to perform within 10% of simulation.

The requirements are met by using a well-known and tested kinematic structure and varying different parameters until optimal performance was reached. The kinematic structure is basically the same as that of a Unimate Puma560, or very similar to many industrial models. Using a standard kinematic form means that there are readily available solutions of inverse kinematics and dynamics. The main difference between our robot and most industrial robots is merely the power-to-mass ratio, as we have almost 5 kW to move approximately 10 kg.

Construction details

The arm is constructed with Amtec PowerCubes. These are self-contained actuators with onboard controllers that can be controlled over a CAN bus. Although the modules documentation claims that they support 1 Mbit/s bus speeds, allowing approximately 6000 messages per second over the bus, practical tests have shown that this causes an overload of the onboard CPU, so that in practice communication has to be limited to sending about 2000-3000 messages per second - which allows for 500-700 Hz control as 4 messages (send current, receive status response, poll velocity, receive velocity) are needed per control loop iteration. Apart from this, testing so far indicates that the modules perform according to specification, with regard to speed, power and precision. The control loop is currently running at 625 Hz.

The arm is mounted on a sturdy (rated for loads up to 1000 kg) industrial table that has been bolted to the lab wall. The table has in turn been reinforced with a 6 mm steel plate. This has so far proved to be a rigid enough mounting.

Power to the arm is supplied via 3 Cosel PBA-1500F units that supply each of the PR 110 units with 48 V up to 30 A for actuation, and one Cosel PBA-1000F unit that supplies 24 V up to 45 A for control electronics and actuation of the PR 070 and PW 070 units. Here is a circuit diagram for the power source.

The following parts have been used for the arm itself:

Part Model no. Comment
1st joint PowerCube PR110 51:1 reduction gear
1st link PAM104 55mm cylindrical rigid link
2nd joint PowerCube PR110 101:1 reduction gear
2nd link PAM108 200mm cylindrical rigid link
3rd joint PowerCube PR110 51:1 reduction gear
3rd link PAM119 45mm conical rigid link
4th joint PowerCube PR070 51:1 reduction gear
4th link PAM106 200mm cylindrical rigid link
5th,6th joint PowerCube PW070 2DoF wrist joint
Dimensions drawing
The D-H parameters, using J.J. Craig's notation.
iαi-1ai-1diθi
10 m0 mθ1
2-90°0 m0 mθ2
30.310 mθ3
4-90°0 m0.51 mθ4
5-90°0 m0 mθ5
690°0 m0 mθ6

Simulation and control Code

Code for simulation and visualization is available. Control code is coming soon.

Source code

The code directory is not yet complete, but will be extended as soon as I have spare time to clean up more of the code and check it for bugs.

Images and video clips

Click on the thumbnails to see larger image.

a picture of popeye
a picture of the
robot arm setup
a picture of the workespace
a diagram of the
reachable workspace

This clip shows one of the first tests of high speed motion with the manipulator, moving 60 cm vertically from standstill to standstill in approx 0.5 s.

Clip of fast position control (4.6 MB)

The following clips show the first prototype implementations of autonomous ballcatching. Catch rate is at present approximately 90% for benevolent throws.

Short clip of ballcatching (6.5 MB)
Longer clip of ballcatching (16 MB)

The following clips shows the first prototype implementation of teleoperated ballcatching.

Clip of teleoperated ballcatching (3.1 MB)


Operator's VR view (5.0 MB)

The following clips show the first prototypical teleoperation experiments with the robot arm. In the clips, the robot is limited to approximately 20% of full performance for safety reasons.
MPEG clip of teleoperation with 0.2 s network delay (32MB)
MPEG clip of teleoperation with 0.5 s network delay (33MB)

Summary specifications

Total reach 0.91m
Max payload 5 kg
Max end effector velocity 7 m/s
Max end effector acceleration 140 m/s2
Max power consumption 5 kW
Control bus CAN (4 channels)
Control frequency1 600/1200 Hz
External connection UDP/IP
Control space Joint/Cartesian
Control types Position/Velocity
  1. The control frequency depends on the velocity measuring strategy. If velocity is interpolated from position readings, only half the number of messages are needed. However, the uncertainty in the time-stamps of the position readings means that the accuracy deteriorates considerably. The control frequency is expected to improve as the control system is more finely tuned.

Publications

  1. Christian Smith and Henrik I Christensen.
    A Minimum Jerk Predictor for Teleoperation with Variable Time Delay
    (accepted for) IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2009). pp XXX-XXX, 2009
    bibTeX abstract

  2. Christian Smith and Henrik I Christensen.
    Wiimote Robot Control Using Human Motion Models
    (accepted for) IEEE International Conference on Intelligent Robots and Systems(IROS 2009). pp XXX-XXX
    bibTeX abstract

  3. Christian Smith and Henrik I Christensen.
    Constructing a High Performance Robot from Commercially Available Parts
    (Accepted) IEEE/RAS Robotics and Automation Magazine (2009)
    bibTeX abstract

  4. Christian Smith, Mattias Bratt, and Henrik I Christensen.
    Teleoperation for a Ballcatching Task with Significant Dynamics
    Neural Networks, Special Issue on Robotics and Neuroscience, vol 24, issue 4, pp. 604-620, May 2008.
    bibTeX abstract Science Direct Pre-print

  5. Mattias Bratt, Christian Smith and Henrik I Christensen.
    Minimum Jerk Based Prediction of User Actions for a Ball Catching Task
    IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2007), pp. 2710-2716, 2007 bibTeX abstract

  6. Christian Smith and Henrik I Christensen.
    Using COTS to Construct a High Performance Robot Arm
    IEEE International Conference on Robots and Automation (ICRA 2007). pp 4056-4063 bibTeX abstract pdf

  7. Mattias Bratt, Christian Smith and Henrik I Christensen.
    Design of a Control Strategy for Teleoperation of a Platform with Significant Dynamics
    IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2006). pp 1700-1705, 2006 bibTeX abstract pdf

The research presented on this webpage is funded in part by the 6th EU Framework Program, FP6-IST-001917, project name Neurobotics.
Disclaimer
Valid HTML 4.01!
Last update: 2009-02-14