Another year is now behind us and with it a decade of the most rapid advances in robotics. It is amazing how much the robotics field has advanced forward in the last 10 years. We only started documenting this process 3.5 years ago but the bases for the current state were put in place in the dieing years of the 90s. In this article, I will attempt a brief review of the most important robotics advances in the last 10 years and in the process try to predict what the current decade may have in store for us.
Before we continue, I want to make one thing clear. Many of the recent developments in robotics can be directly attributed to the increase in computer speed and the abundance of cheap and massive storage. I remember that when I entered graduate school in 1999, the fastest computer we had just purchased was an AMD Athlon chip running at 1GHz with half a GB of RAM. For the same amount of money today, we can purchase Quad Core Intel CPUs running at 3+ GHz with additional processing power provided by massively parallel albeit simple hardware in the form of a GPU. Today, using a series of such GPUs, Scientists with a small budget spend only a few thousand dollars to build extremely fast and useful number crunching machines (see the FASTRA II “supercomputer” for example.)
The faster computers allow researchers to develop advanced probabilistic algorithms that today only take a few hours to execute; in the past, similar algorithms required weeks if not months to finish. Many such algorithms could not even be run to completion due to the limited amounts of RAM available on older computers.
In terms of storage, just a week ago, I purchased 1TB of external storage for a bit more than $100. Moreover, most desktop computers can now be purchased with 2-4GB of RAM for less than $2000. Having access to such high capacity Hard Drives and RAM helps in many applications especially those that require massive amounts of data for learning.
Sonar sensors were the most dominant in the 90s due to their low cost and low data rate which was suitable for processing in real-time using the computers available then. Unfortunately, sonar is too inaccurate for any serious robotics application especially in field robotics, i.e., outdoors applications. In the early part of the last decade, laser sensors and cameras started replacing sonar on robots. In the second half of the decade, lasers took over while cameras were a close second.
In the past, I have written about the importance of solving the Simultaneous Localization and Mapping Problem (SLAM) in robotics. Navigation is the most fundamental skill that a robot should posses. Faster computers and algorithmic advances allowed researchers to build robots that can navigate dynamic environments with ease for days at a time without human intervention. The late nineties saw the birth of a number of museum tour guide robots with CMU’s Minerva being the most successful one utilizing a nice idea from computer vision, i.e., the Condensation algorithm, for localization. This new Monte Carlo algorithm was later extended for solving the complete SLAM problem essentially kick starting a new research field known today as Probabilistic Robotics.
It was in these early years that the SICK LMS laser sensor also started to become very popular. Laser can measure the distance to obstacles with much higher accuracy compared to sonar or stereo vision; the latest model SICK LMS 291 can measure the distance to obstacles 30 meters away with an accuracy of 10mm. The main disadvantage of the laser rangefinder is that it only allows scanning on a single plane so some obstacles such as walls are easily seen but others such as tables not (with laser only the legs are detected.) However, researchers took full advantage of laser sensors and developed advanced SLAM algorithms making it possible to construct extremely accurate floor plans of buildings using one or more fully autonomous robots. The laser rangefinder became a vital component of every vehicle that competed and successfully completed DARPA’s Grand and Urban Challenges.
Today, laser sensors are prevalent in robotics with smaller versions already in the market, e.g., the Hokuyo laser rangefinders, and at a price point that enables their incorporation into small consumer robots such as the Neato robotic vacuum.
As computer speed continued to increase over the years, cameras also became a viable sensor for robotics. Stereo vision is often utilized in domestic robots for localization, object recognition, people recognition and tracking. A stereo camera cannot provide distance measurements that are as accurate as laser for objects far away from the robot but it is great for objects that are within the robot’s grasping distance. In a few more years and as the computer vision community continues to develop new and better algorithms, cameras are destined to become the dominant sensor in robotics.
In the 90s the only serious humanoid robot around was Honda’s ASIMO which had cost millions of dollars to develop; at the same time, research in legged locomotion was certainly undergoing in many labs around the world and especially at MIT’s Leg laboratory. In the last few years, legged robots have taken their fair place as a main vehicle for robotics research. Just about every robotics laboratory in Japan and Korea has its own humanoid robot. Sony developed and eventually discontinued QRIO one of the nicest humanoid robots I have ever seen. In addition, Sony’s robot dog AIBO became a hit for a while until it was also discontinued but not before getting its own league in RoboCup. Just a few months ago, Toyota bested all others by releasing information about their running humanoid robot which is the newest member of the Toyota Robot Partner program. Of course, there are many other humanoid robots such as HUBO, WABIAN-2R, Hoap-3, REEM-B, etc.
Boston Dynamics wowed the world with their BigDog four-legged robot and its smaller brother LittleDog. Japan imagines humanoid robots will be used in space exploration in the not so distant future. In just a couple of years, a small French start-up developed a small humanoid robot named Nao with much less funding than Honda; Nao is now available for sale to academics for just a few tens of thousands of dollars.
Legged robots are also available to hobbyists for just a few hundred dollars. We have written about numerous humanoid robot kits that entered the consumer market in the last 3 years, including Jo-Zero, Bioloid, and i-SOBOT.
In the next few years, I have no doubt in my mind that we are going to be seeing even more humanoid robots developed and made available to consumers. These new robots will be highly skilled with more processing power and longer battery life. We may not enjoy the company of humanoid robots at home for at least another 20 years (my prediction) but we are making fast progress towards that goal.
Exoskeletons and medical prosthesis
The most rapid and unexpected developments of the last decade were in the construction of exoskeletons and prosthetic limbs for the disabled.
No doubt the DARPA-sponsored Sarcos advanced exoskeleton for military use is the most advanced of its kind; for the military, Lockheed Martin has also developed the Human Universal Load Carrier (HULC). The Japanese have focused more on civilian applications and for the first time ever and for a few thousand dollars a month people in Japan can actually rent the HAL exoskeleton. Honda introduced the Honda Legs lower-body exoskeleton which they hope will be of value to factory workers.
Roboticists have also developed extremely dexterous and easy to use arm and hand prosthesis for the disabled. The advanced robotic mechanisms (read about Dean Kamen’s DARPA sponsored Luke Arm and the i-LIMB Hand by Touch Bionics) coupled with new advances in Neuroscience allow users to control them with precision using only their thoughts. These devices will no doubt improve the quality of life for many people especially as their cost goes down over time and they become more widely available. Trials with real subjects have shown this to be a promising and life changing area of research.
In 1997, NASA landed and operated on Mars their first robot rover named Sojourner. However, it was not until 2004 when the twin rovers Spirit and Opportunity landed on the surface of the Red planet and made numerous scientific discoveries over a period of 5 years; the two robots are still operating on the planet. NASA is already preparing a new robot destined for Mars named the Mars Science Laboratory. At the same time, they are preparing the ATHLETE rover for a return mission to the Moon.
In the meantime, even private industry has entered the game of space exploration encouraged in recent years by the establishment of the Google Lunar X Prize (read about the Astrobotic Red Rover). Several international teams have registered and are now competing for millions of dollars in prizes; its team’s goal is to be the first team that lands a robotic rover on the Lunar surface, travels at least 500 meters and transmits live video back to Earth.
Robots will continue to be a crucial component of space exploration for many more years. Robots can be designed to operate for long periods of time in the harsh environment of space while being cost effective in all aspects from construction to mission control and monitoring. I mentioned earlier that Japan envisions robots used to construct space stations on the Moon and other planets in preparation for the arrival of humans. This is not as far fetched as it sounds.
Needless to say, the biggest moneymaker of the last decade was consumer robots. For the first time in human history, we can easily purchase one or more useful and entertaining robots. The best known player in this space is, of course, iRobot and the Roomba robotic vacuum which at the cost of just a few hundred dollars found its way into the houses and hearts of millions of people. The company has followed up with more household robots for floor washing, pool cleaning, and gutter cleaning. The very successful Roomba has been cloned by several other companies hoping to earn a small slice of the vast consumer robotics market.
In recent years, Robotic toys have also grown in numbers and popularity. WowWee has developed and successfully marketed a large number of legged and wheeled robot toys including Robosapien, Femisapien, Rovio, Tribot etc. Lego redesigned their Mindstorms robotic kit and received lots of love from the open source robotics community. Others such as Ugobe entered the toy market promising advanced AI wrapped in a small and cute package, i.e., the robot dinosaur Pleo, but failed; it is expected that in any market some companies will succeed and others will fail and Ugobe belonged to the latter category.
At the same time, lots of humanoid robot kits have also come out of Japan. There are too many to list but the Bioloid kit, Jo-Zero and i-SOBOT are the 3 that come to mind at the moment.
I expect that many more consumer robots will become available during the next few years including more toys that will be easy to program and extend (a hacker’s paradise.) The consumer robotics market is currently only at its infancy and it is very likely going to be the next big industry especially now that the Internet is old news and we are all starting to get bored with Facebook and Twitter.
In the few paragraphs above, I have only discussed a small number out of all the groundbreaking advances in robotics that we have witnessed in the last decade. For example, I have not talked much about robot-assisted surgery in which the da Vinci Surgical System is tops or the numerous advances in Neuroscience, brain-machine interfaces, and therapeutic robotics. Recently, researchers have also been able to simulate large neural networks using large supercomputers; I bet that we are going to be seeing more of this in the years to come.
Open source robotics has also flourished not only through Lego’s Mindstorms NXT kit which I mentioned earlier but also through the massive effort undertaken by Willow Garage.
As fast, power efficient, many-core processors become available in the next few years and with continuing developments in algorithms, robotics will become a large industry we are all going to benefit from. Investing in an education relating to robotics today is probably a good idea in my opinion.
Using the comments section, I encourage you to write your opinion on the greatest decade in robotics. I would really like to know what you think.