Session 2: Next Level Industrial Robots
Intuitive programming, increased mobility, flexible adjustment to the task at hand: These are the characteristics robots will have in the future. They will feature adaptive control based on AI and Machine Learning. A remote prospect? Not quite. Initial use cases already exist. The experts of Session 2 have been and still are involved into such development projects.
These scientists present the upcoming two generations of industrial robots:
Industrial robots will be programmed using new intuitive tools in the future. Prof. Dr.-Ing. Torsten Kroeger, former head of the Google Robotics Software Division and current CTO at Intrinsic, USA, explains what these tools are exactly. Intrinsic is part of the Alphabet group and focuses on developing robotics software and AI applications.
Michael Hartmannsgruber is VP of the Festo robotics team that has developed the innovative Festo Cobot. He presents a collaborative pneumatic robot that better integrates with humans than the robots preceding it, and is accepted just like a human “employee” or aide. Artificial intelligence and machine learning as well as smart pneumatics are prerequisites for making this work.
Dr. Jeremy Wyatt, Director of Applied Science at Amazon Robotics, Germany, presents innovative use cases involving smart robots for handling goods in the world’s largest online retailer’s locations. The core competencies of Amazon Robotics include the development of collaborative and autonomous robots using AI and machine learning for applications deployed across the group of companies.
Ryan Gariepy, co-founder and CTO of both Clearpath Robotics and OTTO Motors, outlines his perspective on the robot market based on fifteen years of robot development experience across various applications. He feels that robotics companies need to change the way they think: Robots are becoming increasingly cost-effective. And they will become easier to operate. New robotics companies will enter the market and rise quickly as they build on existing robot platforms. Some of them could turn into global market leaders very quickly.
Speakers and presentations in this session
Robotics can be defined as the intelligent connection between perception and action. Adding cameras and sensors to robot arms or mobile robots makes the task of robot programming challenging (and often economically non-viable). I will show a few examples of how non-expert robotics application developers can use computer-vision and machine learning algorithms without writing a single line of code or any knowledge about training and inference. Compared to robot perception, creating utility value through machine learning for robot actuation is more challenging - especially when physics plays a role in your application (e.g., dealing with unknown or changing friction, modeling contact forces, unknown and changing process parameters, dynamically changing environments). While much less generic, I will show examples of how a novice application programmer can implement applications using force/torque-controlled industrial robots. I will demonstrate how these concepts can be applied across applications and to a large extent agnostic to robot and sensor hardware using a new software platform.
Torsten is Chief Technology Officer at Intrinsic.
He is the founder and former CEO of Reflexxes GmbH, a startup working on research and development of real-time motion generation software. In 2014, Reflexxes was acquired by Google, where Torsten became the Head of a Robotics Software Division. This included coordinating robotics and machine learning research activities between DeepMind, Google Research, Boston Dynamics, and X.
He is also a Co-founder and former CEO of Loom Vision GmbH. Loom Vision's focus was on software and certification of machine learning and robotics software. In 2018, the team of Loom Vision GmbH joined X.
From 2017 to 2022, Torsten was a Professor of Computer Science at Karlsruhe Institute of Technology (KIT).
Torsten has been working as a research consultant for Volkswagen AG, KUKA Roboter GmbH, Connyun GmbH, Manz Automation AG, Auris Surgical Robotics, Inc., Redwood Robotics, Inc., and Google, Inc.. Torsten is a board member at Fischerwerke and Pictet Asset Management.
Torsten is an editor or an associate editor of multiple IEEE conference proceedings, books, and book series, and the Multimedia Editor of the Springer Handbook of Robotics.
Among other awards, Torsten received the 2022 IEEE RAS George Saridis Leadership Award, the 2018 IEEE RAS Distinguished Service Award, and the 2014 IEEE RAS Early Career Award. He is an IEEE Fellow.
Perception and Actuation in Robotics: Machine Learning Made Simple
Robotics can be defined as the intelligent connection between perception and action. Adding cameras and sensors to robot arms or mobile robots makes the task of robot programming challenging (and often economically non-viable). I will show a few examples of how non-expert robotics application developers can use computer-vision and machine learning algorithms without writing a single line of code or any knowledge about training and inference.
Compared to robot perception, creating utility value through machine learning for robot actuation is more challenging - especially when physics plays a role in your application (e.g., dealing with unknown or changing friction, modeling contact forces, unknown and changing process parameters, dynamically changing environments). While much less generic, I will show examples of how a novice application programmer can implement applications using force/torque-controlled industrial robots.
I will demonstrate how these concepts can be applied across applications and to a large extent agnostic to robot and sensor hardware using a new software platform.
Hightech Summit Session 2: Next Level Industrial Robots
Nature gives us important and valuable impulses for technology. Neuronal networks are copies of nature’s incredible brain design, and we even created a new disciplinary called bionics - the science of copying and adapting nature’s solution not only but also for industrial automation. As one of the emerging fields in robotics are “robots for human” so called cobots – machines that can interact and collaborate with the human. The goal must be to design such machines in a way that they intrinsically accepted by the human companion. Therefore, one focus must be to build-in an intrinsically compliance, another focus must address the easy – or naturally – interaction between human and machine using more sensors and artificial intelligence technology.
The Festo Cobot – pCobot – is such a design approach for such an industrial co-worker. We will show how we use state-of-the-art controlled pneumatics technology to achieve a more “human behavior” without losing the required capability of an industrial machine. To control such a machine is a challenge and here the use of ai and ml is also evaluated. Also important for “robots for humans” is the ease-of-use – especially in unstructured environments - and here ai and ml is also the key to lower the entry hurdle for using cobots. The ”driver assistant systems” for robots will lead us the way to a more and more autonomous system that will be one day not just an robot but more a companion to the human. “Bin picking” is an excellent example for simplifying the setup and operation and shows already the advantage of ai and ml to extend the limits of today’s automation. This talk will show a different design approach for industrial cobots and will also covers why ai and ml is a key technology for robotics – whether to create a more naturally human robot interaction or to increase the usage of robots in service or industrial applications.
Michael is passionate about any kind of Technology with close to 25 years of experience in Industrial Automation and a deep knowledge in various types of industrial robotics. He studied at the University of Stuttgart and at the Karlsruhe Institute of Technology and holds Diploma degrees in Engineering and Business Engineering. Michael has been working for Festo SE & Co. KG in various management positions. Before becoming head of R&D at Customer Solutions, he was leading Festo’s Global Marketing and some smaller Business Fields. As Head of R&D at Customer Solutions he was responsible for the development of cartesian and delta kinematics and various customized robotics projects. He is now part of the dedicated Festo Robotics team, where he leads a highly talented robotics business team with the aim to enter the collaborative robot market with an innovative pneumatic driven robot – the “Festo Cobot”. The “Festo Cobot” was shown at the Hanover Fair and the Automatica fair in 2022 and received a lot of interests by the robotics community and potential customers as well. He loves to develop and play with Technology also in his short spare time.
Hot AIr, Cold Steel
Nature gives us important and valuable impulses for technology. Neuronal networks are copies of nature’s incredible brain design, and we even created a new disciplinary called bionics – the science of copying and adapting nature’s solution not only but also for industrial automation. As one of the emerging fields in robotics are “robots for human” so called cobots – machines that can interact and collaborate with the human. The goal must be to design such machines in a way that they intrinsically accepted by the human companion. Therefore, one focus must be to build-in an intrinsically compliance, another focus must address the easy – or naturally – interaction between human and machine using more sensors and artificial intelligence technology.
The Festo Cobot – pCobot – is such a design approach for such an industrial co-worker. We will show how we use state-of-the-art controlled pneumatics technology to achieve a more “human behavior” without losing the required capability of an industrial machine. To control such a machine is a challenge and here the use of ai and ml is also evaluated. Also important for “robots for humans” is the ease-of-use – especially in unstructured environments – and here ai and ml is also the key to lower the entry hurdle for using cobots. The ”driver assistant systems” for robots will lead us the way to a more and more autonomous system that will be one day not just an robot but more a companion to the human. “Bin picking” is an excellent example for simplifying the setup and operation and shows already the advantage of ai and ml to extend the limits of today’s automation. This talk will show a different design approach for industrial cobots and will also covers why ai and ml is a key technology for robotics – whether to create a more naturally human robot interaction or to increase the usage of robots in service or industrial applications.
Hightech Summit Session 2: Next Level Industrial Robots
Amazon has challenging robotic manipulation problems in terms of scale, item variability and process variability. In the past four years we have invented our own approaches to a variety of pick and place tasks. I will talk, at a high level, about the visual perception, grasp learning, failure detection, continual learning and automated A/B testing that we have used to deploy hundreds of robots that pick millions of packages every day. I will also cover the even more challenging problem of manipulating the hundreds of millions of products we have in our catalogue, including fast pick and place, damage prevention and stowing to dense storage.
Jeremy Wyatt is Director of Applied Science at Amazon Robotics. He leads a team that invents and deploys robotic manipulation systems to handle packages and products. The robotic systems which he has contributed to currently pick tens of millions of packages every week to fulfill customer orders. He was previously Professor of Robotics and Artificial Intelligence at the University of Birmingham, in the UK. He has authored 120 papers on topics in Robotics and Artificial Intelligence, including machine learning, robot vision, robotic manipulation, probabilistic AI and robotic task planning. He earned his PhD from the University of Edinburgh.
Robotic Manipulation at Amazon
Amazon has challenging robotic manipulation problems in terms of scale, item variability and process variability. In the past four years we have invented our own approaches to a variety of pick and place tasks. I will talk, at a high level, about the visual perception, grasp learning, failure detection, continual learning and automated A/B testing that we have used to deploy hundreds of robots that pick millions of packages every day. I will also cover the even more challenging problem of manipulating the hundreds of millions of products we have in our catalogue, including fast pick and place, damage prevention and stowing to dense storage.
Hightech Summit Session 2: Next Level Industrial Robots
From vacuums to quadrupeds to self-driving cars, robots are becoming increasingly physically capable, intelligent and cost-effective. As with any emerging industry, the earliest innovators didn't have the luxury of decades of fundamental knowledge and best practices available to them. They built from the ground up and learned the hard way what not to do. Today, we're entering a new era of robotics. The most successful robotics companies of the next decade won't be the ones building from scratch. They'll build on existing platforms that have been hardened to solve very specific problems, including problems in autonomy, fleet management, simulation, and more throughout the robotics stack.
In this presentation, the audience will learn how robotics development has been done recently, what is changing, and what is coming in the next decade from an expert with fifteen years of experience in robot development & deployment across a variety of industries. Market expectations surrounding robotic capabilities, security and privacy, and robustness and safety are becoming increasingly difficult for new entrants to match. Nevertheless, a variety of market forces are making building robots cheaper and easier than ever before, and demand for robotics has never been higher!
Just as a new software company today wouldn’t build their own cloud computing platform, and instead would use AWS, the next generation of robotics companies are not going to start with a hodgepodge of ROS nodes and custom circuit boards. It is highly likely that some of the world’s largest robotics companies haven’t even been founded yet!
Ryan Gariepy is co-founder and CTO of both Clearpath Robotics and OTTO Motors. In addition, he serves on the board of the Open Source Robotics Foundation, is a co-founder of ROSCon, and also co-founded and co-chairs the Canadian Robotics Council. Ryan is also an advisor to several startups and venture capital groups, and helped found the Next Generation Manufacturing Canada initiative. He is a regular speaker, panelist, and expert guest on topics including robotics, AI, and technology policy. Ryan completed both a B.A.Sc. degree in Mechatronics Engineering and a M.A.Sc. degree in Mechanical Engineering at the University of Waterloo, and has over seventy pending patents in the field of autonomous systems.
Don’t Just Build Robots, Deliver Results
From vacuums to quadrupeds to self-driving cars, robots are becoming increasingly physically capable, intelligent and cost-effective. As with any emerging industry, the earliest innovators didn't have the luxury of decades of fundamental knowledge and best practices available to them. They built from the ground up and learned the hard way what not to do. Today, we're entering a new era of robotics. The most successful robotics companies of the next decade won't be the ones building from scratch. They'll build on existing platforms that have been hardened to solve very specific problems, including problems in autonomy, fleet management, simulation, and more throughout the robotics stack.
In this presentation, the audience will learn how robotics development has been done recently, what is changing, and what is coming in the next decade from an expert with fifteen years of experience in robot development & deployment across a variety of industries. Market expectations surrounding robotic capabilities, security and privacy, and robustness and safety are becoming increasingly difficult for new entrants to match. Nevertheless, a variety of market forces are making building robots cheaper and easier than ever before, and demand for robotics has never been higher!
Just as a new software company today wouldn’t build their own cloud computing platform, and instead would use AWS, the next generation of robotics companies are not going to start with a hodgepodge of ROS nodes and custom circuit boards. It is highly likely that some of the world’s largest robotics companies haven’t even been founded yet!
Hightech-Summit Session 2: Next Level Industrial Robots
