Thursday, April 2, 2009

In-Person Review of Readybot

Pros: A unique approach to mobile service robots the yields a potential breakthrough in cost-to-performance and safety. Control paradigm could work for large-scale deployment. The first service robot design that seems commercially viable in the near term.

Cons: Still a prototype. Control software needs more testing and development.

Introduction

When I first saw the Readybot design in the kitchen-cleaning videos, it seemed interesting but not very sophisticated…a boxy robot on wheels with two primitive arms attached. I was a bit more impressed after seeing the unit at the Robo-dev show last year. Then I got a call from a friend who had seen a live working demo of Readybot at the Homebrew Robotics Club in Silicon Valley…he said “wow…people were really excited, we’ve been waiting years for something like this to come along.”

So I called Readybot, and with a bit of persuasion, I was finally invited to their small office. I came away very impressed indeed. As a long-term advocate for robotics, it’s been a long hard wait for somebody to show me a new concept. Readybot gave me that new concept.

A Different “Take” on Mobile Robots

The most important thing about Readybot is that it’s intentionally simple. Other mobile manipulative robot projects use a traditional articulated robot arm attached to a mobile base. Or in other cases, they build a fully humanoid robot that attempts to mimic human form. These designs are attractive and sophisticated looking, but I have a hard time believing that they can be produced in a practical form for a low enough cost.

According to the builders, Readybot was designed from scratch just for mobile 2-armed service applications. The arms and the body act as a single system, useless without each other, but working together in a way that eliminates redundant joints and redundant cost. After seeing it I came away suspecting that the gold standard for the robotics industry – the 6 DOF articulated robot arm with shoulder pitch and yaw – may be redundant and unnecessarily complex on a mobile robot.

Also the device is quite plain. It’s a white box with 2 arms, more of a small home forklift than Rosie the Robot. Yet with this appliance-minimalist approach works well.

The version I saw was intended as a “trainer” unit, said the developers, which they hope is so simple that other robot builders will be able to clone it.

Goal: 1/10th the Cost

The team impressed on me that their overriding focus is on building a robot that is inexpensive. They believe that the only way to launch the future robotics industry and to create a tenfold increase in the size of the robot market is to create entry-level robots that can be purchased by small business and consumer uses

I’ve heard this kind of talk from other people. But the Readybot people make this speech while having at their sides a robot that actually does what they are talking about. It IS a cheap easy to use personal robot. That adds a certain weight to the argument.

There are 3 areas that I looked at in particular: Base mobility, arms, and control system.

Mobility

The Readybot mobile base is a set of inexpensive gear motors hooked to industrial omni-wheels. It uses the Northstar sensor (Evolution Robotics) to determine location in the room, with infrared distance and ultrasonic sensors for fine-tuning and obstacle avoidance. It worked well in a set of 3 test rooms.

Of course this has none of the self navigation that is present in other mobile robot bases from firms such as Mobilerobots Inc. I asked how they could justify a mobile base that can’t navigate around obstacles. I got the following reasoning:




>>It's designed to be upgraded. The Readybot arm/body assembly could be removed and attached to a Mobile Robots Pioneer or Patrolbot bases, or other commonly available self-navigating bases. “If people want sophisticated SLAM then no problem, just upgrade the base” said Tom Benson, the product designer, “but let’s give them something cheap to get started with, that works in a simple environment.”

>>The collaborative cloud control system (which I describe below) gives robot programmers the ability to use the remote human operator as the “obstacle avoider of last resort.” Essentially the robot can switch to a tele-operated mode any time needed.


This gives a very good flavor of the overall design process for Readybot. They are willing to accept less autonomy in return for lower initial cost, but are architecting the system to allow later upgrades to autonomy if desired by the owner.

Cheap, Strong Arms and Base, Virtual Shoulder Joints

The Readybot robot has two arms that can carry about five pounds each (they said, upgradable to 10 pounds with a simple change in gear ratio). These arms are primitive by robotic standards, but again, very inexpensive. Benson said that the arms were the most important component in the whole project: “Two arms on a mobile base, that can reach table top height, are the key to the whole robotics field from now on. We were convinced of that from the start. The problem was to make them cheap and safe.”

The Readybot solution is clever. They are simple extendable linear arms on a rotating shoulders with a vertical lift (otherwise known as a cylindrical robot arms) mounted on each corner of the device, so the two arms have full access to the front, side, and rear of the robot.

Again these are much simpler than the more common articulated arms used in other designs. They have no shoulder joint and no elbow joint, which eliminates the most complex and expensive parts of the arm, and greatly increases the effective weight capacity. The lack of a shoulder and elbow joint creates significant reductions in arm dexterity however this is compensated by the mobile base, which has omni-wheels and can move side to side. This allows the mobile base to act as a “virtual shoulder joint” giving the robot additional range of motion to make up for the loss of the shoulder and elbow.

Intentionally Slow Moving, Imprecise, But Safer Joints

Most robotic designs have stiff, uncompromising, and tightly controlled joints which require powerful motors and gearboxes. They are designed that way intentionally in order to provide high speed, repetitive, and precise placement. This also makes them dangerous since the stiffness requires high-inertia gearboxes that give the robot a tremendous “punch.” Get in they way of a moving industrial robot and you may end up injured or dead.

The Readybot design abandons this entire concept. Their arms are slow moving, relatively sloppy, and in some planes of motion quite weak.

How can this possibly work? Benson’s comment was “a mobile robot that moves from place to place on wheels is inherently imprecise anyway. Trying to put precise arms on an imprecise base is misplaced effort and limits your design options”.

In the Readybot design, the joints that rotate, or swing the arm have almost no torque; only enough to free-swing the 5-lb lifted load. This makes them much safer. But if you want to open a door or push an obstacle aside, don’t you need some strength in the arm to do so? “Not in the swing axis” said Benson “in fact we’ve found in many cases the optimum solution is to allow the arm to swing completely free. Then we orient the robot facing diagonally to the door and pull with the linear extension part of the arm, which is quite strong. It pulls the door handle towards itself and the free swing of the rotating axis allows the door curvature to move freely as it is pulled.“

Collaborative Cloud

Their control software use what they call a "Collaborative Cloud", which is used to control the robot in day-to-day work.

This is a mixture of tele-operation and scripted behavior. The operating system and application platform for the robot is built with the assumption that the robot is connected to a broadband connection at all times. Via that connection, the robot communicates with a “cloud” of remote servers with scripting, set up data, and processing support. And more importantly the robot communicates with, and is controlled by, a “cloud” of human supervisors. What if that internet connection is broken? “The robot stops instantly”.

Benson says this was inspired by work at Nasa, CMU, Idaho National Labs, and a number of other places, all of whom have been working on different types of collaborative control and adjustable autonomy.

The plan is for human supervisors to control many robots at a time by using a customized library of script elements and routines to “build programming on the fly” as the robot works through its day. These supervisors could be on-site or in remote locations. For example a manufacturing facility might have 50 robots run by 4 operators in a local control center. Or 1000 robots taking care of elderly folks in their homes might be run by 100 contractors working from home.

Again, this cuts costs because most complex processing is offloaded to the cloud thus the robot needs little on-board processing. Instead of multiple “cores” on the robot, which is very common for other designs, this one uses a single computer.

Caveats and Quibbles

They made it clear that the system isn’t ready for commercial use yet; version 1 is a prototype, version 2 is under construction. The Readybot team has been working on a tight budget, so their design will continue to be less finished compared to others (for example Willow Garage, who as we’ve discussed earlier, has been generously funded and which I'm eager to see).

But for an old-timer like myself, these quibbles are beside the point. Readybot has a smart, determined team. They have made fundamental advances in cost-performance and control systems. At this early stage such cost reductions are crucial because they indicate that the service robotics market as a whole can be expected to follow the same cost-cutting trajectory of other successful high tech markets.

2 comments:

火之河 said...

learns a lot,many thanks!

Joey1058 said...

I enjoyed your review very much. it was very objective and informative. Thanks!