|
|
Tenzing: A Dartmouth
|
|
|
|
|
|
|
Tenzing: A Dartmouth
|
|
|
|
|
Tenzing is a rock climbing robot built by Dartmouth Computer Science and Engineering students for CS88/188 in the spring of 2004. The robot is named in honor of Tenzing Norgay Sherpa the first climber on top of the world's highest peak. The robot uses computer vision to find handholds and to localize the robot on the wall.
Tenzing uses eight hobbyist servo motors to control the four elbow/knee joints and four shoulders/hip joints. It is designed to climb a flat wall with randomly placed hand holds. Each hand/foot has a force sensor that is used to control the pressure on each hand and the body has a tilt sensor that helps keep the body upright. All the low level coordination of the robot is done by an embedded HC12 controller. The HC12 communicates with a Java-based server running on a PC using a RS232 serial port. The server uses a camera and image processing software to find the position of the robot on the wall and to locate hand holds. The robot can climb in and interactive mode where a person uses the computer mouse to click on a sequence of handholds or AI software can automatically plan a path up the wall.
|
These pages have pictures taken during the construction
and original programming of the robot for
CS88/188 |
We are working on series of movies that show Tenzing with different levels of intelligence. As all you roboticist know getting your simulation code to run on a mechanical robot can be difficult.
| Teleoperated
teleoperated.avi
|
Robot is teleoperated from a Java console. Messages are sent via a RS232 serial port to robot to control motors and read measurements from force sensors and tilt sensor. Force sensors are used to maintain pressure on hand holds. This movie is speed- up x8. |
Visual Servoing
|
Limbs are now controlled by dragging a stick
figure in the GUI. Visual servoing is done using a web cam and LEDs at the
hands and feet. You will see the image darken after the arm/leg is
dragged. This is done to increase the visibility of the LED that we are
servoing on.
In the second video, instead of using points generated by the path planning algorithm to grab handholds, Kwasi, specifies way points for the arm path by clicking on the screen. These videos are not speed up. |
Path planning for Arm
|
Using the GUI a hand/foot is dropped on a new hold. A* is used to find the best path to the new hold. Visual servoing is then used to guide the hand to just above the hold, and then force-feedback is enabled for the limb and the hold is grasped. |
Standing up
|
Robot autonomously moves from one stance to
the next and stands up. The sequence of hand moves were not determined
autonomously, but in the case shown in the video determining the sequence
of hand holds to use is quit trivial.
|
| Full Path Planning |
These videos are compress using 3ivx MPEG-4 codec. You can download codecs for Windows , Mac OS and Unix/Linux.
We have several simulation to demonstrate the path planning algorithms used by Tenzing. Click on the images to run a Java-based simulation.
Arm Path Planning and Standing Up  |
This applet
shows
Their is a bug where the feet don't always find a path to the hold even though the hold is reachable. |
| Full Path Planning (coming soon) |
This applet shows the best path up a wall. |
(coming soon)
The Jet Propulsion Lab's (JPL) Mechanical and Robotic Technologies Group has built a rock climb robot named Lemur. Timothy Wolfe Bretl at Stanford's Aerospace Robotics Laboratory implement a stochastic path planning algorithm for Lemur. This MPEG-1 video shows the robot climbing open-loop up a climbing wall.
Case Western Reserve's Biorobotics Lab built a cockroach-like robot that uses
force feedback to keep its posture.
Here are some movies:
narrated introduction and
moving around
Michigan State University built a
suction-cup wall-climbing robot.
