Omnidirectional personal transport platform

- Ken Perlin


With all the buzz around Dean Kamen's Segway personal transport device (left), I figured I might as well jump in and talk about my own experiments in the general area of personal transport devices.

My ideal personal vehicle is a small platform (say, around 15 inches in diameter), sort of a hi-tech magic carpet, which lets me glide around, moving and turning at will, simply by shifting my weight. It would also compensate for my weight shift to stop me from falling over. Unlike the experience of riding a Segway, I'd like to be able to navigate simply by shifting my weight and get rid of the handlebar entirely, so that I can keep my hands free.

You could also really use such a device if you were a paraplegic, or wanted a low-cost and very maneuverable omnidirectional traveling chair. In that case, you would operate a joystick with one hand: tilting the joystick to make your chair travel in the direction of the tilt; twisting the joystick to rotate.

So what kind of machine would let me do this which is simple, robust, lightweight and inexpensive? What follows is a fairly high level outline of one approach I've been looking into.

The components:

The platform rests on three equilaterally spaced wheel-pairs. Each wheel-pair consists of two small wheels, closely spaced next to each other on a common axle. Each wheel-pair is able to spin freely about the vertical axis between the two wheels of that wheel-pair. Each wheel is driven by its own small electric motor, so all together there are six small motors. The use of six small motors, rather than, say, two bigger motors, allows the vehicle to have a very low profile.

The rider's shifting weight is sensed by (at minimum) four force sensors around the perimeter of the vehicle body: front-left, front-right, rear-left, and rear-right. From the weight at each of these four sensors, we can tell three key things: (i) how much the rider is leaning forward or backward, (ii) how much the rider is leaning left or right, and (iii) how much the rider has placed one foot further forward than the other. For example, if the front-left and rear-right sensors collectively have more weight on them than do the front-right and rear-left sensors, this indicates that the rider has placed his left foot further forward than his right foot.

The vehicle responds to leaning by traveling in the direction of the lean. For example, leaning forward makes the vehicle travel forward, and leaning backward and to the right makes the vehicle travel backward and to the right.

Also, the vehicle responds to one foot forward by rotating: if the left foot is placed more forward than the right foot, then the vehicle rotates to the right. Conversely, if the right foot is placed more forward than the left foot, then the vehicle rotates to the left.

The reason that we need at least four sensors is that the vehicle must respond to three degrees of freedom: (i) how fast the rider wants to travel forward or back, (ii) how fast the rider wants to travel left or right, and (iii) how fast the rider wants to turn clockwise or counterclockwise. We can get one degree of freedom from the first two weight sensors, and then one more for every additional sensor we add.


Each of the three wheel pairs is able to travel by rotating its two wheels in the same direction, and to quickly spin around (thereby changing the direction of travel) by rotating its two wheels in opposite directions. Different combinations of travel and spin are achieved by driving the two wheels differentially.

A microprocessor, attached to the center of the underside of the platform, converts the loads sensed at the four weight sensors into varying positive and negative voltages, which are amplified to drive each of the six motors. Power is supplied by batteries attached to the underside of the platform. The batteries are located around the perimeter of the underside of the platform, in the spaces between the wheel-pairs.

The Java applet below shows the principle of the platform in operation, as it might be seen from below, looking up through a transparent floor. The copper colored rectangles are the motors; the black rectangles are the wheels. The red dots show the locations of the weight sensors. The blue rectangles indicate the positions of the battery packs.

Click here to see a 3D model of the vehicle, in which I experiment with a round vehicle shape.


How powerful should the motors be?

There are six motors. If each motor is has about 60 watts of power, then the vehicle will have 360 watts of total power, or about 1/2 horsepower. Good examples of appropriate kinds of DC motors can be found at: Igarashi Motors.

What kind of gearing should it have?

Each motor has a gear on its shaft one inch in circumference, which directly drives one wheel of the platform. If each motor spins at about 9000 rpm, then the platform travels, at peak speed, at about 12000 inches per minute, or about 12000 × 60 / (12 × 5000) = 12 miles per hour.


Once I began working into this, I eventually discovered that someone else had already come up with the idea of using an equilateral triangle of independently steerable wheels:

Mobile Robot Capable of All Directional Movement: The robot is designed for an autonomous motion in a partially known environment with some dynamically changed conditions. Local collision-avoidance and sub-goal handling can be performed with fuzzy if-then rules. Genetic algorithm (GA) is applied to optimise the mobile robot paths planning. The three wheel vehicle is designed so, that axes of wheels hold angle 120 and wheels are placed on the peaks of equilateral triangle. This disposition has an advantage against the four wheels design, that all three wheels are all the time in a contact with a base (road), what is given by the fact, that a plane is defined by the three points. Directions and speeds of individual wheels are independent. Combination of rotation directions and speeds of individual wheels provide translation and rotation movement at the same time.

Contact: Dr. Pavel Osmera, Professor
Tel.: +420 5 4114 3332

In addition, I found that some researchers had also developed the idea of using "split casters" for the wheels:

Wada, M. and Mori, S. (1996). Holonomic and omnidirectional vehicle with conventional tires. In Proceedings of the 1996 International Conference on Robotics and Automation, pages 3671-3676, Minneapolis, Minnesota. IEEE.
which some other researchers at MIT then did follow-up work on:
Yu, H., Dubowsky, S. and Skwersky, A. (1999). Omni-directional Mobility Using Active Split Offset Castors.


Given the combination of the above work, together with the various new things I'm bringing to it, such as:

there's still a lot of innovation and work to be done, particularly in making the whole thing lightweight and inexpensive, (not more than 20 lbs, less than $500 retail), and getting the control loop well tuned and intuitive, which is mainly a software problem. Ideally, when you're standing on it, the moving platform would just feel like an extension of your body, as though you're simply floating to where you want to go.