September 15, 2009      

With the successful completion of a second milestone test for its PR2 mobile robot platform, Willow Garage demonstrates its  understanding of crucial issues, and it significantly advances the development of personal, mobile robots.

In June 2009, Willow Garage, a developer of hardware and open source software for personal robotic systems, announced that its prototype Personal Robot Mark 2 (PR2) completed its second milestone by navigating through eights doors and plugging itself into nine separate power outlets over the period of an hour.

Willow Garage did not issue a formal press release announcing the completion of its milestone—the company is, after all, a research company heavily populated with engineers and academics, and it’s not naturally inclined to writing formal press releases. The story was picked up, however, by a number of robotics blogs, science-focused Web sites, and the general press (most notably the New York Times). Most of the articles were short, simple, and descriptive, with the follow-on comments in typical Web fashion—that is, even shorter, simpler, and less descriptive (although the 1960s TV show The Jetsons and R2D2 were mentioned). The stories reported what occurred during the second milestone test, but failed to answer the “So what?” question.

Robotics Business Review believes the PR2 milestone carries much more weight than can be discerned from a cursory reading of brief articles and blogs. With the completion of this second milestone, Willow Garage demonstrated that it understands many of the fundamental issues that must be addressed prior to commercial-class, personal mobile robots becoming a reality. A fuller understanding of the implications of the Willow Garage announcement will serve to benefit robotics system developers, integrators, investors, and anyone else with an interest in developing a market for personal mobile robots.

The Company
Willow Garage was founded and funded as a research lab in late 2006 by Scott Hassan, one of the designers of the original Google search engine in the ’90s. Hassan was also founder of eGroups, a group email messaging company that Yahoo Groups purchased in 2000 for $450 million. In addition, he has worked as a software architect/developer of Alexa Internet, a Web information company, as well as at the Stanford Digital Library. Bottom line: Hassan is a tech guy, with a good deal of start-up and business experience—not to mention deep pockets. Hassan serves as Willow’s chairman of the board.

Willow’s President and CEO Steve Cousins joined Willow in May 2007 after 20 years of holding senior management and research positions at the IBM Almaden Research Center in San Jose, Calif., and at the Advanced Systems Development Laboratory at the Xerox Palo Alto Research Center.An additional 35 or so employees, most of whom are researchers and engineers with a background in robotics and artificial intelligence, have joined Hassan and Cousins—many of this roughly 35 are SRI International and Stanford University alums.

Willow Garage is not a commercial enterprise in the traditional sense (i.e., products and services sold at a profit). Originally, the company worked in a number of different areas, including autonomous surface vehicles and unmanned cars as well as robot hardware and software platforms. The company then decided to limit its focus and resources. Willow’s current raison d’être is to develop standard robot hardware platforms that use open source software and distribute them to university research groups in an effort to speed the develop¬ment of robotic applications, commercial robotic systems, and breakthrough research.

While Willow Garage is not under the gun for driving revenue through the sales of its platforms, it has said it will apply for patents for hardware when it deems appropriate. In addition, if certain hardware technologies become successful, they could be spun off as separate companies.

The First Milestone
Willow Garage reached its first milestone in December 2008. At that time, its PR2 mobile robot—named after Lord of the Rings’ wide-ranging wizard, Gandalf—autonomously traversed π (3.1415) kilometers of the Willow Garage offices two days in a row.

The company announced that the armless Gandalf had been doing a 3.1415-kilometer run for approximately two weeks before the robot was able to string together two days of uninterrupted travel. During that two-week period, Willow Garage engineers improved the robot’s navigation software and eliminated software bugs. Willow Garage indicated that following completion of the first milestone, it would continue to develop and refine the robot’s hardware and software, as well as extend the system’s navigational capabilities.

While the first milestone might seem somewhat arbitrary—and indeed may be arbitrary in design—it can nonetheless be seen as a test of robustness and durability as well as a facilitator of iterative, navigational improvements. Whether or not Willow Garage specifically designed its systems for continuous use, that fact is that future service robots will run more or less continu¬ously and autonomously. This stands in stark contrast to the bulk of today’s service robots, most of which are teleoperated or programmed to run for relatively short periods of time.

Industrial robots, which are constantly at work, are programmatic (as opposed to teleoperated or autonomous) and operate in highly structured environments. Willow Ga¬rage’s first milestone was a preliminary, tentative step toward creating mobile, autono¬mous, service robots for continuous operation in unstructured environments (more on unstructured environments below).

Sensing, Movement, Control, and Manipulation
The second of PR2’s milestones required that the system finds its way around the Willow Garage office for 26.2 miles (the distance of a marathon). As part of the second milestone, the PR2 navigated Willow’s offices, locating 10 electrical outlets and plugging into nine of them to recharge its batteries. The PR2 system took four days to runs its marathon, and just under one hour to perform its plug-in circuit.

Like the first milestone, the successful completion of the initial segment of the second milestone was a further confirmation of the PR2’s ability to navigate autonomously (sense, compute, resolve, and move), running continuously if need be. During the plug-in circuit phase of the second milestone, the PR2 again demonstrated its sensing capabilities, but this time while also controlling and manipulating its 7-degrees-of-freedom arm and end effector.

At the most basic level, autonomous mobile robots perform only two tasks: moving without supervision, and interacting with their surrounding environment in order to perform a task. While the first milestone addressed only the “move” component of the autonomous mobile robot behavior (sensing and movement), the second test addressed the “act” aspect of behavior (control and manipulation).

Failure Signifies Success
The second milestone trial was notable for its failures as well as its overall success. Willow Garage engineered the second milestone so that it tested for both partial and complete failure. The test was designed so that the robot was required to plug into 10 different designated outlets, although only nine of the selected outlets were physically reachable (a door was locked). Also, the doors throughout the space were variously closed—either partially closed or fully open.

According to Willow Garage reps, the robot also had difficulty grasping its plug and opening a partially open door, but after a number of unsuccessful attempts, it wasable to perform both tasks. In the case of the locked door, the PR2 demonstrated full failure recovery (the door cannot be opened) and was able to resume its circuit. Instances of complete and partial failure, as well as degrees of recovery, are common in our daily interaction with the unstructured world around us. In this early test, the PR2 was able to demonstrate that it too could react and recover from varying degrees of failure.

Semistructured Environments
The building in which the second milestone test was carried out was designed to conform to the accessibility standards given in the Americans with Disabilities Act (ADA). All hall¬ways had a stable and regular surface, and all doorways lacked round doorknobs and had a minimum clear opening of 32 inches with the door opening 90 degrees. This approach simplified the robot’s task, but it also illustrates one design consideration for future robots and their working environments.

The description above suggests that PR2 is not designed, in the literal sense, to operate in fully unstructured environments. The second milestone took place in a weakly unstruc¬tured, yet standardized, working environment—a “semistructured environment,” for lack of a better term. This attribute of the milestone test (and again, the decision may be arbitrary on Willow’s part) illustrates an important, yet rarely discussed, aspect of future mobile robots. That is, it is often better to modify the working environment for mobile robots than to design robotic systems to operate in fully unstructured environments.

For industrial robots, workspace modification is the norm, while for other classes of service robots, such as military ground systems, search and rescue robots, sewer inspec¬tors, and the like, there is no choice but to operate in highly unstructured environments. For many robot applications, however—and perhaps the majority of future systems—it makes better sense to slightly modify the working environment to suit the robot than to overengineer the system to operate in more unstructured workspaces. This approach, which again would be suitable for many applications, would act to reduce systems costs while increasing system functionality and reliability.

Power Autonomy
As with all classes of mobile robotic systems, power is a major gating factor to expanded use and autonomy. In most cases, the battery life of commercial mobile robotic systems is only a few hours to, perhaps, a day. To reach their full potential, such systems should be expected to be mobile for weeks, months, or even years.

Most efforts at increasing mobile robot runtimes involve the development of lightweight, long-lasting, high-density power supplies and/or proprietary charging systems. But even the best battery systems run down and must be recharged. Dedicated charging stations provide one solution, but being proprietary they are, well, proprietary, and that implies a limited number of them—not to mention the necessity of reducing a given system’s operational range so as to be in close proximity to a charging platform. What is required is the availability of a charging station wherever the robot happens to be. But how might this be possible?

To demonstrate the PR2’s ability to manipulate objects in its working environment, Willow could have settled on some highly photogenic task that required advanced manipu¬lation capabilities, such as playing a piano or holding a wine glass. Instead, however, the company had the robot locate standard power outlets and plug itself in.
The second milestone demonstrates that in standard, indoor environments, the PR2 has, essentially, a charging station anywhere it might find itself (and, therefore, an unlimited power source). The robot’s operational area is not limited by proximity to a charging station; standard power outlets are everywhere. This class of autonomy—that is, power autonomy— is necessary, but not sufficient, for the development of truly useful mobile robots.

The Power of Open Source
The first and second milestones tested the stability of the PR2 platform, along with naviga¬tion, failure recovery, fine manipulation, and power autonomy. The tests were also notable in that they demonstrated the robustness of the open source software that powered the two systems.

The Willow Garage team firmly believes in the power of open source software de¬velopment and an open source distribution model to drive advances in robotics. With theopen source model, code is made freely available and may be modified and redistributed, subject to the terms of a licensing agreement.

Unfortunately, depending on the software in question, open source libraries can suffer open source robotics software. It is also from inadequate services for testing and bug fixes, lack of continuous upgrades, and security the primary developer of the open source and reliability problems. To address these issues (while still taking advantage of the open robot operating system (ROS), originally source model), Willow Garage actively participates in the robotics open source community. developed at Stanford University. For example, the company supports the OpenCV vision library and Player/Stage, a TCP/ IP-based middleware for robot device control. Willow is also the primary developer of the open source robot operating system (ROS), which was created in the Stanford Artificial Intelligence Laboratory and is now accredited under a BSD license.

Progressively Addressing Issues
The successful completion of the second milestone by Willow Garage’s PR2 system demon¬strates the quality of the company’s mobile robotics research platform and, more important, an understanding of the key obstacles and issues that must be progressively addressed before personal mobile robots can enter the mainstream. Those issues include:

  • Mobility, Localization, and Navigation—the ability to navigate environments autonomously
  • Manipulation—the ability to use fine control to manipulate objects in its environment
  • Failure Response—the ability to automatically recover from an error state
  • Power Autonomy—the ability to overcome power source limitations
  • Open Source—increasing the strength and quality of open source robotics software
  • Working Environments—the practical necessity of designing robots to operate in standardized, semistructured environments
The Bottom Line

  • Willow Garage is well funded, led by seasoned management with both engineer¬ing and business backgrounds, and it is largely populated with robotics and AI researchers and engineers from Stanford University and SRI International.
  • Willow Garage is chartered with building research robots based on open, standard hardware and software technology and distributing them to research institutions.
  • Willow’s PR2 is a mobile, autonomous robot platform designed for continuous, autonomous operation in unstructured environments.
  • The PR2’s first milestone was a test of robustness and durability as well as a fa¬cilitator of iterative, navigational improvements.
  • The PR2’s second milestone tested sensing, movement, control, and manipula¬tion as well as various levels of failure and recovery, the types of interactions commonly encountered when operating in the “real world.”
  • It is often better to modify the working environment for service robots than to design robotic systems for operation in fully unstructured environments.
  • Modifying working environments to more easily accommodate the operation of mobile robots reduces development costs while increasing system functionality and reliability.
  • Batteries and proprietary charging stations place limits on mobile robotics systems.
  • Willow Garage tested the manipulation capabilities of the PR2 by having it plug into standard power outlets, increasing the range of its mobile platform.
  • Willow Garage supports many types of open source robotics software in addition to being the primary developer of the open source robot operating system (ROS) originally developed at Stanford University.