Challenge 1: Robots must become as skilled as humans
Humans are extremely dexterous. They produce fluid, expressive, and complex movements. They can handle delicate objects effortlessly. Humans also have the ability to develop new skillsets by training and refining these movements. Take, for example, professional puppeteers. Through repetition and application, they bring marionettes to life. Utilizing creative skills, they imagine how best to recreate a movement that accurately portrays a particular character or an animal. In close collaboration with ETH Zurich, computer animators and graphic designers working with Disney Research Zurich closely study the movements of people and animals in their natural environments, in order to recreate realistic scenes and animate lifelike characters. Robots, on the other hand, still have quite a long way to go before they can match the level of dexterity and skill that we see in people. Yet, if robots are to be useful in the real world, society expects robotic technologies to perform on par with human-like capabilities.
Currently, my team in the Computational Robotics Lab studies complex human motions, like the example of the puppeteer, to understand how we can program robots to grasp the physical behaviors of such systems. By imagining how to animate marionettes robotically, we can model some of these motions and determine how to optimize movements. At the ETH Zurich Pavilion in Davos, Switzerland during the 2020 World Economic Forum’s Annual Meeting, we debut the PuppetMaster—a physics-based motion-planning framework for robotic animation of marionettes. We also reveal our mathematical model that forms the basis of a general unified framework enabling optimization algorithms to infer concurrently a robot puppeteer’s workspace, movements, control, and dynamic motions.
Challenge 2: Robots will change the world of work, but to the benefit of society
Why are we interested in robots that can recreate human movements? The reason lies in potential real-world applications that might free up human labor for more skilled jobs, such as in hospitals or hotels where robots might be able to clean rooms and make beds. Mail sorting facilities with their automated pipelines are another example. In today’s modern postal facilities, letters and boxes fly through robotically assisted sorting centers at great speed. However, the changing nature of shopping and the advent of online offers across the world means that even the shape of packaging is evolving. Instead of standard shaped envelopes and boxes, mail systems now deal with soft and oddly shaped packages as well.
Today, mail centers still need to hire people to deal with non-standard packaging, as opposed to installing robots. When a crumpled package needs to be picked up, unfolded, and placed back on a conveyor belt with the label face up, today’s robots have neither the ability to react to such a situation, nor the skills to deal with it. We are a long way off from the day in which robots will be able to manipulate such objects—even in jobs that appear to be simple and repetitive like mail sorting, cleaning, or making beds. Our challenge in the end is to make sure we find a balance. One day, as they become increasingly skilled, robots will be able to improve quality of life for people who require assistive care, for example. Smart robots will be able to ensure that our society operates efficiently, allowing creative jobs that require more skills and experience to remain in human hands. Mundane or dangerous tasks will go to robots, and overall that will improve society.
Challenge 3: Robots must learn how to work alongside other robots and people
Robots excel at performing identical and repetitive jobs. While robots can move with super-human skill, strength, and precision, they can only repeat the precise motions for which they were programmed. In contrast, even at a young age, children have the ability to adapt to a very dynamic environment with a multitude of other actors. Adaptation is something that robots are very bad at doing. We are trying to overcome such limitations in our research at ETH Zurich. We simply do not have the technology to instill human-like reasoning and interaction in robotics technologies—at least not at this time. As soon as they are required to react or adapt to changes in their environment, robots are just not up to the task. This is why the foundation of our work starts with a robot like the PuppetMaster, because this already requires an advanced level of understanding of the physical behavior of complex everyday objects, such as marionettes.
Considering how robots interact with each other and with humans forms the basis for advancing robotics research to the next level. We see a huge potential for human-robot collaboration in resolving complex tasks more efficiently in the future. The groundwork we lay with the PuppetMaster will enable us to move on to challenges that are more aspirational.
These are just a few of the incredible challenges that inspire our daily research. My research team aims to endow robots with human-level dexterity when it comes to manipulating complex physical systems. We want to give robots the ability to process a problem and come up with a suitable solution. We have already achieved, to some extent, the ability for a robot (Skaterbot) to teach itself how to skate. In the long-term, we anticipate creating robots with a capacity to predict reactions, to exercise a certain degree of creativity, and to come up with optimal responses. By drawing inspiration from what people do in everyday activities and endowing robots with these abilities, our society stands to benefit greatly. Smart robots are not something we need to fear, rather, if we do it the right way, it is something that we should embrace. In due course, robots will be capable of handling complex objects, interacting with actors including other robots and people, and evolving to adapt to everyday situations. Let the robot-learning journey begin!
a global affairs media network
Three Challenges in Rendering Next Generation Robots
January 24, 2020
Setting the record straight, Stelian Coros deconstructs some common challenges in advancing robotics. He reveals the realities and challenges of robotics research, and offers an exciting outlook on the evolution of robots with human-like capabilities.
Challenge 1: Robots must become as skilled as humans
Humans are extremely dexterous. They produce fluid, expressive, and complex movements. They can handle delicate objects effortlessly. Humans also have the ability to develop new skillsets by training and refining these movements. Take, for example, professional puppeteers. Through repetition and application, they bring marionettes to life. Utilizing creative skills, they imagine how best to recreate a movement that accurately portrays a particular character or an animal. In close collaboration with ETH Zurich, computer animators and graphic designers working with Disney Research Zurich closely study the movements of people and animals in their natural environments, in order to recreate realistic scenes and animate lifelike characters. Robots, on the other hand, still have quite a long way to go before they can match the level of dexterity and skill that we see in people. Yet, if robots are to be useful in the real world, society expects robotic technologies to perform on par with human-like capabilities.
Currently, my team in the Computational Robotics Lab studies complex human motions, like the example of the puppeteer, to understand how we can program robots to grasp the physical behaviors of such systems. By imagining how to animate marionettes robotically, we can model some of these motions and determine how to optimize movements. At the ETH Zurich Pavilion in Davos, Switzerland during the 2020 World Economic Forum’s Annual Meeting, we debut the PuppetMaster—a physics-based motion-planning framework for robotic animation of marionettes. We also reveal our mathematical model that forms the basis of a general unified framework enabling optimization algorithms to infer concurrently a robot puppeteer’s workspace, movements, control, and dynamic motions.
Challenge 2: Robots will change the world of work, but to the benefit of society
Why are we interested in robots that can recreate human movements? The reason lies in potential real-world applications that might free up human labor for more skilled jobs, such as in hospitals or hotels where robots might be able to clean rooms and make beds. Mail sorting facilities with their automated pipelines are another example. In today’s modern postal facilities, letters and boxes fly through robotically assisted sorting centers at great speed. However, the changing nature of shopping and the advent of online offers across the world means that even the shape of packaging is evolving. Instead of standard shaped envelopes and boxes, mail systems now deal with soft and oddly shaped packages as well.
Today, mail centers still need to hire people to deal with non-standard packaging, as opposed to installing robots. When a crumpled package needs to be picked up, unfolded, and placed back on a conveyor belt with the label face up, today’s robots have neither the ability to react to such a situation, nor the skills to deal with it. We are a long way off from the day in which robots will be able to manipulate such objects—even in jobs that appear to be simple and repetitive like mail sorting, cleaning, or making beds. Our challenge in the end is to make sure we find a balance. One day, as they become increasingly skilled, robots will be able to improve quality of life for people who require assistive care, for example. Smart robots will be able to ensure that our society operates efficiently, allowing creative jobs that require more skills and experience to remain in human hands. Mundane or dangerous tasks will go to robots, and overall that will improve society.
Challenge 3: Robots must learn how to work alongside other robots and people
Robots excel at performing identical and repetitive jobs. While robots can move with super-human skill, strength, and precision, they can only repeat the precise motions for which they were programmed. In contrast, even at a young age, children have the ability to adapt to a very dynamic environment with a multitude of other actors. Adaptation is something that robots are very bad at doing. We are trying to overcome such limitations in our research at ETH Zurich. We simply do not have the technology to instill human-like reasoning and interaction in robotics technologies—at least not at this time. As soon as they are required to react or adapt to changes in their environment, robots are just not up to the task. This is why the foundation of our work starts with a robot like the PuppetMaster, because this already requires an advanced level of understanding of the physical behavior of complex everyday objects, such as marionettes.
Considering how robots interact with each other and with humans forms the basis for advancing robotics research to the next level. We see a huge potential for human-robot collaboration in resolving complex tasks more efficiently in the future. The groundwork we lay with the PuppetMaster will enable us to move on to challenges that are more aspirational.
These are just a few of the incredible challenges that inspire our daily research. My research team aims to endow robots with human-level dexterity when it comes to manipulating complex physical systems. We want to give robots the ability to process a problem and come up with a suitable solution. We have already achieved, to some extent, the ability for a robot (Skaterbot) to teach itself how to skate. In the long-term, we anticipate creating robots with a capacity to predict reactions, to exercise a certain degree of creativity, and to come up with optimal responses. By drawing inspiration from what people do in everyday activities and endowing robots with these abilities, our society stands to benefit greatly. Smart robots are not something we need to fear, rather, if we do it the right way, it is something that we should embrace. In due course, robots will be capable of handling complex objects, interacting with actors including other robots and people, and evolving to adapt to everyday situations. Let the robot-learning journey begin!
