The fourth and final day of RIA’s Robotics Week began with a roundtable discussion between experts in research and artificial intelligence (AI) technologies. AI has been incorporated in many devices in many industries. Everything from household pressure cookers to massive manufacturing equipment has the capacity for different levels of AI, and these advancements are changing the way we think about how robots can be used. These innovations have exploded in the last five years and will continue to do so.
“Everything has changed with the complexity of our challenges with robotics,” said Jon Battles, director of World Wide Robotics Advanced Technology at Amazon. “Robots are moving into consumer products, they’re moving into agriculture, you name it…Those applications have become more and more complex.”
With more complexities comes the need for more tools to address those challenges, which Battles said is sometimes confused with actual AI technology, which includes vision systems and advanced sensors. In order to develop these tools that will give more power to AI, data has to be gathered. And where better to start that process than with Industry 4.0?
“Industry 4.0 was all about connectivity, and when the connectivity started happening, there was tons of data from the sensors, the robotic sensors, the cameras,” said Murali Gopalakrishna, head of Product Management, Intelligence Machines at NVIDIA. “All of those brought perception into robotics.”
Juan Aparicio, senior director of R&D Engineering at Siemens, discussed researching and implementation of AI systems. Many times, technologies like AI are researched but cannot scale up for larger applications outside the lab setting.
“You can always optimize,” he explained. “You’re controlling a very complicated electromechanical device that also influences the environment as it’s moving.”
The human-in-the-loop concept is necessary for the advancement in AI robotics. As these systems are placed in increasingly more complex environments, they will get stuck and will require a human hand for aid.
“Augmented control and human-in-the-loop are also becoming a really important factor for enforcement learning,” said Battles.
Like many technologies, AI-powered robotics is better used in some areas than others. In manufacturing, robotics is commonly used in bin picking, material handling, grasping, pick-and-place operations and digital simulation.
But what’s really next for AI?
“We’ll just go back to the U.S. robotics roadmap,” said Battles, referring to an earlier Robotics Week presentation. “Everything I’ve seen from new silicone to new advances in neural net performance and direct edge…we’re going to continue to make great progress on all of these fronts.”
A lot of AI performance comes down to computing chips—those that are specially designed for advanced robotics.
“We have purpose-built robotic computers,” said Gopalakrishna. “We are not using a chip that was built for something else…it’s not about having a chip and having it compute. You need to have access to different aspects of the robotic sensors in different ways.”
The panelists agreed scalable systems that enable cloud computing, and that the digital twin will play large roles in the future of AI technology. One of the prerequisites for all these network-based aspects of AI is having the infrastructure that allows for it, whether it’s the cloud or edge computing. But there isn’t just one answer to that.
“Depending on the use case—what you’re going to be using it for—will define the architecture,” said Andrey Shtylenko, global AI solution architect at Honeywell. “The shorter cycle you have from collecting the data and fine-tuning your algorithm, the better for you.”
The panelists agreed that the neuromorphic engineering, data values, infrastructure and broader use cases for AI will drive AI-powered robotics into the future.