Matt's Homepage

Intro

This page has been created in order to make public information that may be useful to other people with research interests similar to my own. Things you will find here include information on artificial self-replicating machines, legged robots, and a number of links I find interesting. This stuff is what I work on in my spare time. During my un-spare time I am employed as an engineer with the robotics division of General Dynamics Land Systems.

Kinematic Self-Replicating Machines

I am interested in building artificial systems that can create copies of themselves. This is a potentially intractable or potentially trivial problem to solve, depending on how you define it and who you're talking to. The answer may be a nanoscale assembler, a computer virus, a cellular automaton, a robotic lunar colony, a test tube containing a population of self-catalytic (or cross-catalytic) molecules, a capable rapid-prototyping device, a field-programmable gate array, a robot wandering a warehouse full of robot parts...

I am primarily interested in solving this problem with a macroscale, but not immense, machine (think desktop size). Sometimes these are called clanking replicators. Over the years I have approached this problem by biting off chunks of it that are small enough that I think I can solve them, yet complex enough that the solution is somewhat interesting. The pictures above came from research I did as a mechanical engineering student at the University of New Mexico (1999-2001). You can download the thesis for more explanation here:

A Physical Prototype of a Self-Replicating Universal Constructor (MS Word 2.9 Mbyte)

Quadruped Project






Overview

The goal of the quadruped project is to combine many of the useful behaviors that have been demonstrated in prior research walking robots into one small, self-contained platform. Recent research investigating the way animals move, and the creation of robotic vehicles based on these investigations, have shown that effective, life-like mobility can be demonstrated by even very simple systems if their mechanics and control devices are properly designed and coupled. This design is in the medium stage. Detailed solid models and schmematics have been created, fabrication methods identified, and bench-tests with hardware have been conducted to verify design concepts (in other words, the robot hasn't been built yet).

The dimensions are roughly 18"L 12"H 11"W, with a weight target of about 7lb (46cm X 30cm X 28cm, 3.2 kg). Each leg has 3 active degrees of freedom. The most important two of these control position of the foot in the sagittal plane of the robot. The third degree allows the leg assembly to be steered as though it were an automibile wheel. This arrangement allows the two strong motors of each leg assembly to act together when propelling the robot in the sagittal plane, such as when moving forward or climbing. The orientation of the third "steering" degree of freedom allows for smooth turns while walking, and agile maneuvering in close quarters - such as when backing up or turning away from an obstacle. Each leg has one passive degree of freedom, a linear spring in the telescoping segment below the knee. This is almost essential for realizing dynamic gaits. The eight primary motors are high-quality 5-watt DC gearmotors. The four steering motors are as yet undefined, but probably can be hefty RC servos. Basic mobility control will be accomplished via a small microcontroller such as the OOPic, and a collection of custom-made, closed-loop motor control boards.


Key Capability Goals

True Dynamic Gaits
Examples include the trot, gallop, and bound. In each of these gaits the robot becomes airborne for a period of time during every walking cycle.

Fast Movement
The speed goal for the quadruped project is 1+ meters per second, which is relatively fast by current standards. This level of speed is often easily obtained by specialized research robots, but the speeds of most self-contained and/or commercially available walking platforms fall well below this mark.

Simple Control
By careful attention to design of robot mechanics, the computational demand of a walking machine can be dramatically reduced. Examples include the open loop control algorithms that generate dynamic gaits in McGill's SCOUT II quadruped, and the motor-less, computer-less bipeds at Cornell's Passive Dynamic Walking Lab.

Agility
Examples include slow walking and fine control of direction of movement, smooth gait transitions, recovery from rollovers, and ability to sit down and stand up.

Climbing behavior
Examples include crossing gaps and ascending and descending stairways. These behaviors will be controlled using reflexive techniques similar to those shown in CWRU's Robot II and the MIT AI Lab's Ghengis.

Availabilty
The intended method of manufacturing for the majority of the quadruped's mechanical components is polyurethane casting in silione molds (Synair). This is a simple and highly effective technique for producing low-to-medium volume of intricate plastic parts, as described in my thesis work above. Masters for component molds can be created either by conventional machining (Sherline machine tools) of raw chunks of polyurethane plastics, or by rapid prototyping techniques. Any university lab with access to a rapid prototyping machine should be able to produce both working components and mold masters with a rapid prototyper. It is my intention to make available all essential design information necessary to replicate fully functional quadrupeds. This includes fabrication drawings and solid models of mechanical components, electrical schematics, pc board layouts, and source code. I believe this open architecture approach (see Linux and OpenPino) will lead to the most rapid development of the technology.

Other Walkers


This is a collection of walkers I built over a few summers as a student at LANL. They are all controlled by variants of the simple nervous net control system developed by Mark Tilden and others. They're surprisingly capable given the simplicity of their controllers. The large one in back and three small ones lined up in front are all 2-degree-of-freedom machines. One motor drives the front leg segment and one the rear. The two rotationally symmetric quadrupeds are 4DOF, having one motor per leg. The hexapod is 12DOF, using a four-bar linkage and 2 motors per leg. The medium-sized quadruped on the left is 8DOF, having mechanics similar to the quadruped described above, but without the ability to steer the legs.

The most successful machine was the symmetric quadruped on the lower right. It could move at fairly good speed (20cm/sec), could walk forward, backward, and sideways with equal ability, had 8 tactile sensors and 4 light sensors. The behavior was quite simple - the vehicle was attracted to light and repelled from obstacles - yet it could explore complex environments for a fairly long time before getting stuck or confused.

Links

Artificial Self-Replication Page by Moshe Sipper - Lots of information on self-replication research.
McGill University Ambulatory Robotics Lab - Info on SCOUT II, RHex, and other ARL projects.
CMU Robotics Institute - Robots of all shapes and sizes.
MIT Leg Lab - The original hopping and running machines.
Case Western Reserve University Biorobotics Lab - Complex cockroach-inspired walkers.
Cornell's Passive Dynamic Walking - Bipeds that excel while lacking motors and control; plus the best walking robot in the world.
Polypedal Lab - Fantastic biological research with direct relevance to walking robots.
OpenPino - Open architecture biped robot.
Acroname - Excellent hobbyist robot supply.
Solarbotics - Supplies and info on nervous nets and BEAM robots.
The Legged Robot Builder - Research robots of the world, animal gaits, simulation, hardware... Good stuff.
Clock of the Long Now - When was the last time we designed something to last 10,000 years?

Matt Moses

Created on ... August 23, 2003