Dr. Tekin Meriçli is a Senior Robotics Engineer at the National Robotics Engineering Center (NREC) of the Robotics Institute at Carnegie Mellon University. Durtti wants Tekin to share how today’s “next generation” robots are already starting to significantly improve our lives.
In 2009, you gave a speech at Kirikkale University in Turkey about “Robotics Applications of AI”. What, if anything, would you change about that speech now, with the benefit of knowing how technology has actually changed since then, Tekin?
My recurring observation had been that the majority of the robotics events and competitions organized by student clubs at Turkish universities primarily focused on extremely primitive, line follower variant robots, competing in meaningless, aimless, and made-up setups.
The presentation at Kırıkkale University in Turkey was one of the several invited talks I gave on the subject of “intelligent robotics” with the intention of broadening the students’ visions and horizons by informing them about what other cool application possibilities as well as competitions, such as RoboCup and the DARPA Urban Challenge, were out there.
Being a veteran of the DARPA Urban Challenge myself, which essentially ignited all of the self-driving car development efforts we’re seeing nowadays, I could have put slightly more emphasis on the importance of that application.
Another application that I’d briefly touched on, but could have elaborated on a bit more, was warehouse automation, where brilliant AI-driven implementations, such as the one by Kiva Systems/Amazon Robotics, are drastically improving our daily lives by enabling more energy and labor efficient storage and management of goods, resulting in much faster, cheaper, and reliable deliveries.
Also, deep learning hadn’t taken off yet back then; but I’m hoping that the examples I’d given on machine learning applications in robotics inspired some of the students to study that subject.
You have been closely involved in the development of semi-automated robots for the elderly and disabled. What are 3 of the most important considerations when developing a robotics system of this nature?
When it comes to assistive robots, which literally live/work side-by-side with people, safety becomes our utmost concern as the robot has to be physically very close to the user’s body and face while performing some of the activities of daily living (ADL), such as self-feeding and grooming. Hence, the robot has to be extremely reliable, ideally formally proven to be safe and fault-tolerant from both hardware and software design perspectives.
Another important consideration is how to make the robot a natural part or extension of the user. Depending on the level of assistance needed, design and development of suitable, intuitive, and flexible human-machine interfaces play a significant role in the applicability and adoption of such technologies.
Last but not least, as is the case for any other intelligent robotic system operating in the wild, the deployed assistive robot has to be robust to uncertainty present in the real world. In order to be able to deal with the aspects of changing user needs in their dynamic and cluttered environments, the robot has to be equipped with multi-modal mechanisms for perception, prediction, learning, and adaptation.
You are very close to your brother, and you and he share many similar interests in computer technology. What do you remember most about that Sinclair ZX Spectrum that you both shared as children?!
My brother, Dr. Çetin Meriçli, has been a trailblazer, a wonderful role model, and a fantastic collaborator throughout the years. Except for my Master’s degree, we’ve graduated from the same schools, worked with the same co-advisors, been members of the same research groups and teams, co-chaired various academic events including RoboCup 2011 Istanbul event, and co-authored 20 publications.
We are both at NREC now and we still work together on a lot of exciting projects.
The Sinclair ZX Spectrum was our first computer and our vehicle of introduction to the field of computer science in the late ‘80s. Aside from the programming exercises that shaped our way of thinking about how to be creative and design efficient software to run on extremely limited hardware (the computer had 48KB of memory, and its Z80 processor ran at 3.5MHz), we loved playing games on that machine!
One of our favorite games was called “Penetrator”, where we were flying through a cavern a spaceship of sorts that could shoot forwards and drop bombs beneath to prevent missiles from launching and radars from collecting intelligence. Even though it was intended to be a single player game, we’d devised a way of exploiting it by having one of us focus on flying the ship while the other one took care of the weapons part. It was a lot of fun!
Tell us about the purpose of the CHIMP robot that you and your Carnegie Mellon University “Tartan Rescue” team developed for the DARPA Robotics Challenge.
The DARPA Robotics Challenge was inspired by the nuclear disaster that followed a massive earthquake and the resulting tsunami occurred in Fukushima, Japan in 2011, when the general devastation prevented workers from reaching and operating the valves that could have prevented the gas explosion that further damaged the reactors. Given that such an environment would be deadly for humans, semi-autonomous dexterous robots would be the natural candidates to deploy.
CHIMP (CMU Highly Intelligent Mobile Platform) was designed to handle such harsh conditions, and be able to operate vehicles and dexterously manipulate tools and interfaces intended for humans mostly autonomously, taking only high level commands from human operators. Its humanoid form augmented with tank-like tracks on each of the four limbs enabled CHIMP to drive over rough terrain on four limbs, and be statically stable when upright on two limbs. CHIMP’s six cameras tightly coupled with its spinning LiDAR sensors and extremely precise inertial navigation system provided 360 degree situational awareness to the operators. Once a high level command from the operator was received, CHIMP was able to carry out the rest of the task autonomously, which made it robust to variances in the communication bandwidth.
What effect do you think AI will “realistically” have on society in the next 3-5 years?
With personal assistants, recommendation systems, navigation route planners, and even semi-autonomous driving, AI has already become an integral part of our daily lives as individuals. However, some of the recent advances will change the scale of the impact, and as a result, we will start seeing some serious job transformations as well as changes in the way we live our daily lives.
For instance, fully autonomous transportation, some applications of which will likely be achieved and commercially available in 5 years, will start reshaping the entire ecosystem built around assumptions like people owning, operating, and storing the vehicles. Lodging facilities and restaurants along busy trucking routes as well as parking lots and garages in the cities will be underutilized and hence lose their business appeal. Vehicle manufacturers will reshape their businesses around this new concept, resulting in job transformations within and around that ecosystem as well. The entire shopping culture will change with the delivery drones bringing the groceries to our doors.
Aside from such intelligent robotics applications, we will start seeing a lot more AI involvement in medicine and law. Back in the ‘70s, expert systems used to be one of the first successful forms of AI software applied in those domains. With the advances in machine learning, particularly in deep learning, expert systems will make a comeback. Deep learning systems have already started identifying and diagnosing diseases on medical data and imagery better than humans. Also, AI involvement in medicine will accelerate drug discovery process and enable much more effective personalized medicine applications.
Similarly, it takes an AI system a fraction of a second to go over millions of legal cases to identify similarities, draw conclusions, and make better-informed and more accurate decisions.
This means that even the jobs that currently require higher education, like general medical practice and law, will go through significant transformation. We are definitely on the cusp of an AI revolution, and it will be exciting, to say the least, to observe the effects at different scales.
As a passionate RoboCup aficionado since 2004, what’s the biggest challenge to creating a successful team of soccer playing robots, in your opinion?
Although it may not be apparent, playing soccer is actually an extremely sophisticated task, both at the individual player level and at the team level. It requires individual players to have various capabilities, like fast and reliable perception to recognize all objects/features of interest on the field (the ball, the goals, field lines, teammates, opponents, etc.), self positioning within the field based on those perceptions, short and long term plan generation at various abstraction levels, fast, agile, and flexible motion execution, and non-prehensile manipulation to drive and kick the ball in desired directions without grasping it.
On top of that, the players need to coordinate with each other, assume and dynamically change roles (defender, midfielder, attacker, supporter, etc.), and overall act as a team to maximize their effectiveness in controlling the ball and dominating the game.
Each of those components need to be very well designed, implemented, and integrated; therefore, the biggest challenge in developing such a team is principled adoption of an end-to-end systems design approach to the problem.
What changes are we likely to see in the way that robotics systems are designed and developed in the next 5 years?
If we consider the design of anthropomorphic systems, my expectation is to see more artificial muscles and soft structures covered with tactile sensors instead of rotary actuators and rigid structures with limited dexterity and tactile sensing.
In addition to the design of general purpose robotic systems, I think there will be a lot more use of machine learning and automated evolutionary design processes to come up with novel special purpose mechanisms.
Finally, Tekin, and with the benefit of your broad experience in the field, if you could pass on just one piece of advice to a brand new class of robotics students, what would it be?
Robotics, by its very nature, is a multidisciplinary field; therefore, an aspired roboticist should try to be a polymath with as much in-depth knowledge as possible in multiple subfields of AI and robotics as well as other disciplines. Having expert-level knowledge in a broad range of domains and disciplines will definitely give you an edge in the field, both in academia and in your business ventures.
Tekin is a member of The Artificial Intelligence Group on LinkedIn.