xBerry Talks – Challenges of SLAM and Autonomy with Piotr Skrzypczyński and Janusz Będkowski
When an interesting and exciting subject meets experts from the field who have some thoughts and knowledge to share – here come xBerry Talks. In this edition of the xBerry Talks series, we mostly discuss SLAM, automation, and the possibilities that a new approach to technology can offer. In this edition, we talked with Prof. Piotr Skrzypczyński, Ph.D., D.Sc. in Robotics, and Janusz Będkowski, Ph.D., D.Sc. in Mobile Robotics. If you’re not familiar with the SLAM and autonomy topic yet, check out our last blog ‘Autonomous robots/cars – why aren’t we there yet?’ by clicking this link.
Autonomy In Housekeeping – Is My Vacuum Cleaner Autonomous?
In our daily life, we are surrounded by numerous machines – some more, some less intelligent than others. Amongst all of them, many of us have equipped ourselves with cleaning robots. Those little machines can make our homes spic and span, but have you ever considered what hides behind their shells? Can we think about our vacuums as autonomous robots, and why is there a lot of talk about vehicle autonomy when autonomy in limited conditions seems like a simpler solution?
“Looking at the current state of knowledge, it is troublesome to make an unequivocal statement about the level of autonomy exhibited by cleaning robots. Simplifying autonomy to the capability to build a map of the environment, navigate this map, and perform a series of missions without the need for an operator, then the answer is yes. On the other hand, many researchers and users of these robots expect spectacular competencies from an autonomous machine. The natural aspiration of the field of robotics is to create a fully intelligent machine that can recognize the environment and, on this basis, perform tasks independently or with the participation of other machines.” says Professor Janusz Będkowski.
“By terming a robotic agent ‘autonomous’ I mean autonomy in the sense of independent decision-making. Thus one can say that an intelligent agent perceives its environment via sensors and acts rationally upon that environment with its effectors. This definition is close to what Russel and Norvig define in their ‘Artificial Intelligence: A Modern Approach’ textbook used widely to teach AI at Universities. In this sense, modern robotic vacuum cleaners are indeed autonomous, although most of them are pretty simple in their behaviours, being reflexive agents rather than more advanced ‘rational agents’, which act towards achieving a goal, in terms of Russel’s and Norvig’s classification. In practice, autonomy may have many aspects, and the absence of some functions (e.g., map building or planning) does not mean that an agent or a robot is not autonomous as long as it makes decisions upon the observations or sensor readings.” claims Professor Piotr Skrzypczyński.
“Why do people talk about autonomous cars rather than about autonomous vacuum cleaners or lawnmowers? Perhaps just because cars are fancier. Driving a car is also an experience most people are familiar with, so ‘robotizing’ this activity looks like moving the autonomy in robotics to the next, important level, unlike operating a vacuum cleaner, which might be perceived as just ‘automatisation’. All in all, the hype about self-driving cars is, in my opinion, mostly a matter of public relations. I like the parallel with the space programme in the 60s. In fact, sending men to the moon was very complicated, risky, and pretty much unnecessary. Still, the entire Apollo programme enabled great progress in many areas of technology, most of them not directly related to space travel. The role of self-driving cars seems to be similar, but nowadays, the driving force is rather from the big tech companies than from the governments.” adds the professor.
Autonomy on Streets and Sidewalks – Approaches and Possibilities
While autonomous driving will undoubtedly happen in the near future, it remains controversial. It turns out that when looking for technological solutions for autonomous robots or cars, we still have to consider a whole range of rules and dependencies related to, for example, the rules of the road. Where will autonomy appear first – on the sidewalks, in the form of last-mile delivery robots effectively manoeuvring between pedestrians, or on the streets, as full-fledged vehicles without a human driver behind the wheel?
“Last-mile delivery robots are perhaps the newest addition to the set of promised robotic solutions we should have at hand in the near future. However, such robots are likely to overtake their larger cousins in the race for real-world applications. There are a number of factors that make the delivery robots both technologically simpler, and easier to deploy in real applications,” – claims Professor Piotr Skrzypczyński. – “First of all, the risks related to any accidents involving food or parcel delivering robots are lower and easier to handle than those related to accidents of autonomous vehicles that can transport people. Of course, driving through a crowd is also not easy, but there are algorithms that can manage such scenarios, while the risk of an unpredicted but serious accident is considerably smaller.” – he adds.
This view is also shared by Professor Janusz Będkowski. “The commercialization and introduction of autonomous food delivery robots into everyday use will be faster than autonomous vehicles for transporting people with the highest level of autonomy (no steering wheel). At the same time, as current cars are partially autonomous, they can already stay in a lane, automatically brake and accelerate. The problem of autonomy boils down to making a comfortable journey from point A to point B, which is a challenge for cars from a safety point of view. Food delivery robots are much lighter and move at a slower speed. You can even risk saying that they are a much easier solution due to the limited impact on other road users. The probability that a food delivery robot will cause a fatal accident is estimated to be significantly lower than in the case of a fast-moving autonomous car,” – claims Professor Będkowski.
Autonomy = SLAM? Let’s Talk About Technological Thought Patterns
When talking about vehicle autonomy, we immediately start thinking about SLAM as well. Is this thought pattern correct? Is building and updating dense and accurate maps necessary to create autonomy?
“Autonomy without SLAM is possible but to a limited extent,” says Professor Janusz Będkowski. “SLAM enables immediate map update, which consequently reduces planning time and eliminates the risk of mission failure from point A to B. Automatic, continuous map update is a challenge. In my opinion, autonomous vehicles of the future will use a map that will be updated on the basis of a large amount of information from various sources. Therefore, robots will use maps from other sources apart from SLAM on the onboard computer.”
“Yes, it is possible,” – claims Professor Piotr Skrzypczyński.“In fact, considering the definition of autonomy and autonomous/intelligent agent we discussed earlier, it is possible to show intelligent behaviour without building any world model. This is something that was discovered in the late 80s when robotic researchers started to question the sense-plan-act paradigm in constructing autonomous robots. This paradigm, taken directly from prominent AI theories at the time regarding knowledge representation and formal planning, turned out to be insufficient for real-time systems in an unpredictable environment, that is, for mobile robots. Rodney Brooks – an influential and now famous robotics researcher, coined the phrase ‘the world is its own best model’, referring to reflexive agents.
So, we do not need SLAM for autonomy. However, the trick is that without a world model, we merely need to stay at the autonomy and cognition/planning level of these simple vacuum cleaners. To plan actions and movements, and to achieve a notion of optimality in the actions taken by a robot, we need to have a world model. In turn, if we want a world model, we need to build it, as the use cases where we can get a reasonable world model from an ‘oracle’, such as HD maps in autonomous driving or architectural blueprints for indoor navigation are not available for many environments, and frankly speaking, their use is rather constrained. If anything, in such a model, the robot needs at least the ability to update it. A model of the environment can be obtained in a relatively simple way as long as the poses of the agent are known, but again we need an ‘oracle’ to provide these poses. Hence, we arrive at a chicken-and-an-egg problem, which is nothing other than SLAM.
SLAM is necessary for autonomy if having a map of the environment is necessary in order to complete the task at hand. Obviously, in some environments, we have a practical ‘oracle’, such as GNSS for outdoor navigation, but all technical systems have their limitations. GNSS technology makes no exception, and due to the limited signal availability and a number of issues related to the propagation of the GNSS signal from the satellites, the SLAM with local sensing is still necessary. The type of map required by a robot also depends on the application. Usually, SLAM can work with a pretty sparse representation of the environment that contains only features useful for location. However, if we want to use the same map for motion planning, object recognition, semantic segmentation or anything else than SLAM itself, then a dense map might be the right choice. However, in my opinion, it makes sense to keep the SLAM map as minimal as possible due to the scalability properties, and then use the agent’s pose estimates provided by SLAM to register any other properties of the environment.” says the professor.
SLAM’s Impact on the Future – Is Autonomous Society a Fact or Fantasy?
In pop culture, it is not difficult to find numerous references to an automated, robotic future. Ideas about robots replacing people in various fields, professional or social, have become a popular narrative in cultural texts with an undertone of science fiction. Does this fiction, often created purely for entertainment, have a chance to be transferred to reality? Maybe we have a chance to create a partially automated society in which we will live in a kind of symbiosis with robots, not only helping us in the most repetitive tasks but also acting almost as everyday companions?
“I don’t think robotics and autonomy will take over the industrial and economic sectors one by one. This is rather an evolution, or silent revolution, as the process has a fast pace, but it is not easy to observe it. We are witnessing an increasing level of automatisation in many sectors, and this is most often intelligent automatisation, based on image processing, estimation, and learning. These processes are matured, and robust enough to be applied whenever they are required almost out of the box, as long as the problem at hand is narrow and well defined.” claims Professor Piotr Skrzypczyński.
Due to the progressing ageing of society, it will be necessary to introduce autonomous nurses, autonomous cleaning robots, autonomous robots working in food production, and others. That will enable, firstly, to reduce the negative impact of the decreasing number of employees, and secondly – to provide healthcare. The set of existing professions will be enriched with completely new professions directly related to the developing technology of mobile robots,” says Professor Janusz Będkowski. “It is easy to imagine the emergence of numerous services providing assistance oriented around maintaining autonomous robots in operational readiness. It is also easy to imagine an almost fully-autonomous world in which man no longer has to work. Fortunately, this is just a utopia – autonomous mobile robots must first meet broadly understood social acceptance. And would that be easy for us? As a possible answer, it is enough to cite the first-ever fatal accident of an autonomous car, which, in my opinion, set back the development of technology by at least a decade. To sum up, autonomy still faces a significant number of challenges, not only technological but also social. Therefore, working on autonomous vehicles is still exciting.
I am sceptical as to the chances to see fully autonomous cars on the streets in the next ten years, or to have robotic butlers at home in the next two or even three decades,” adds Professor Piotr Skrzypczyński.“But on the other hand, the techniques taken from robotics and AI will be embedded into more and more devices we interact with on a daily basis. My personal tips for the next targets of robotization are logistics, retail stores, and agriculture. All these sectors require a large number of low-skilled workers and include rather repetitive tasks. So there we have both the reason (shortage of workers), and the opportunity (repetitive tasks are easy to automatise).
This will, of course, have a societal impact, as not all people can migrate to more creative jobs. Concerns about new AI technologies causing job losses have been around for a long time. While automation driven by the development of AI and robotics is creating new and better jobs for professionals, it is also eliminating jobs for those who are not skilled and creative. This process has accelerated over the past two years as the pandemic has prompted many companies to save money, and the implementation of new technologies reduces costs, increases productivity and reduces dependence on employee availability. The World Economic Forum (WEF) stated in a recent report that AI and robotics-based automation will eliminate around 85 million jobs by 2025.” concludes Professor Skrzypczyński.
Are you curious about how SLAM can benefit your business? Contact us and see how xBerry’s expertise can support you.
Experts bios of this edition:
Professor Piotr Skrzypczyński
Poznan University of Technology
Piotr Skrzypczyński is a full professor at the Institute of Robotics and Machine Intelligence (IRIM) at Poznan University of Technology (PUT) and head of the IRIM Robotics Division. Piotr Skrzypczynski received the Ph.D. and D.Sc. degrees in Robotics from PUT in 1997 and 2007, respectively. He has authored or co-authored more than 160 technical papers in robotics and computer science.
His current research interests include AI-based robotics, robot navigation and SLAM, computer vision, and machine learning.
Janusz Będkowski is D.Sc. and Ph.D. in mobile robotics. He is a researcher at the Polish Academy of Science, Institute of Fundamental Technological Research, and a former engineer at TomTom International BV. He is actively involved in the theoretical and practical aspects of simultaneous localization and mapping applications on a global scale.
Currently, he has been working on a methodology for closing the gap between geodesy and cartography, geoscience and mobile robotics. He is an active member of the European Land Robotic Trial (ELROB) and the European Robotics Hackathon (ENRICH).