Pneumatics offer great advantages in power density, weight, and cost and is the technique of choice for kinematically gifted robots. Air is used to move missile actuators, the MIT/UTAH dextrous hand, most animatronics, and the jumping, running, somersaulting performance of Raibert's Leg Lab robots.

A Spin Off of the Proto-Android Project

Desktop Pneumatic Testbeds

David Santos/ July '93/ Austin, Texas

The Proto-Android is an ambitious39 DOF pneumatic robot underconstructionand promises to be a control challenge.

Therefore two pneumatic development platforms have been lashed up, in order to give the control engineers a head start on mastering control issues while the mechanical team completes gory work on the android carcass itself.

Joe, James, and David Perez, David Hutchings, and David Santos have done the shop work so far. Norman Annal has been drafting Proto-Android components and drafted plans for third party use.

The trial rigs consist of modified computer monitor arms, saving hundreds of dollars. Each arm has 2 DOF's worth of 1 1/2" inch cylinders with a variety of position encoder options. The arms are able to knock your teeth out before you can rightly recollect what happened, so caution is needed in operation.

Various subcomponents, such as pressure sensors and position encoders, are to be competitively tested for their suitability. A nice grab pile of evaluation units have accumulated.

Each arm evaluates a distinct solution to airflow modulation. The cheap, simple, and possibly quite adequate approach is a "digital valve network" where multiple valves give different "speeds".

An expensive and theoretically superior method is to use proportional servo-valves that smoothly transition to any airflow rate, given an analogue input. Dan Novak of Schenck-Pegasus has kindly offered to loan the $750 valves required for one of the arms.

The real-world result for the Proto-Android may be a hybrid digital net, with a few analogue channels rationed to master valves and dextrous subsystems like a hand or a neck.

Its time to start defining the servo-computer interface. The goal is fast, smooth, and reliable reactive control both at the local and global level. For now anyone who wants to hack a quick demo is welcome but a clean long term solution is needed.

This should be a servo protocol standard independent of actuator type, maybe even the Futaba style control signal we have considerable existing code for. Analogue force feedback using pressure sensors is an additional complication over the hobbyist servos we are used to. Classical control theory applies here.

Given adequate control of each test arm they can then be networked for co-ordinated behavior. This will be a simple exercise compared to the combinatorial explosion of the full Proto-Android platform.

Two arms mounted "upside down" on a hydraulic platform as controlled by public interface at Robofest 6.

The "consulting" engineers so far are Geoffrey Schecht, J.P. Mason, Bruce Waters, and Harry Bolch. At the most recent meeting the subject of position feedback was explored and some detailed specifications generated. Programmable Array Logic (PAL) devices will be used to give each sensor, in this case digital position encoders, a unique address on a sensor bus.

Support for this phase of the project comes in part from the City of Austin (Berzerkwerks) whereby the robot arm(s) will do something, however lame, for the public at that event come September.

Warning- Anyone who feels they can contribute to the efforts described is welcome to participate.