Despite the seemingly fantastical demonstration of walking and jumping robots, today’s robots are often stupid, brittle and inflexible, only capable of working in carefully engineered environments, and unable to respond dynamically and sensibly in unexpected circumstances.
Deep learning has been spectacularly successful in certain problems (facial recognition, object recognition, etc.) but the “smart robots” that we have promised still haven’t arrived. The promised robotic future is still a long way off.
New Silicon Valley robotics startup Robust.AI aims, firstly, to build the world’s first industrial-grade cognitive platform for robots. Secondly, it will aim to help companies in a wide range of areas, from construction to eldercare and domestic robots, toward the goal of making robots that are smarter, safer, more robust, more context-aware and more collaborative.
Initially located in Palo Alto, Calif., Robust.AI has secured a “substantial” undisclosed seed round from Playground Global, among other undisclosed investors.
The market for intelligent robots is worth potentially several hundred billion dollars a year, once robots can enter new markets (e.g. construction, eldercare, to-the-door delivery) that historically have been too challenging for traditional robotics.
Robust.AI plans to make money by licensing its cognitive platform, and by helping companies solve robotics problems that would otherwise be out of reach of current technology.
To create this vision, the company has two stellar founders: Rodney Brooks, co-founder of iRobot and Rethink Robotics, co-inventor of Roomba, the best-selling consumer robot of all time, and former chair of the MIT AI lab (CSAIL); and Gary Marcus (CEO co-founder of Geometric Intelligence, acquired by Uber, bestselling author and cognitive scientist at NYU). Brooks will be CTO, Marcus will be the CEO.
Coming from different perspectives, Marcus (cognitive science) and Brooks (robotics) have been writing for the last several years about the perils of deep learning, and why it had been overhyped; they also independently reached similar conclusions about the need for developing machine-interpretable common sense as a prerequisite for reaching the next level of AI.
When Marcus decided to take the plunge into robotics, he says he realized immediately that Brooks, a legend in robotics, would be the perfect collaborator, and Marcus spent months recruiting him. The two say they are excited by the mission and the commercial potential.
“We are building an industrial-strength cognitive platform — the first of its kind — to enable robots to be smart, collaborative, robust, safe and genuinely autonomous, with applications in a very broad range of verticals from construction and delivery to warehouses and domestic robots,” Marcus told me.
“We will be synthesizing a wide range of advances in AI, including both deep learning and classical approaches, with a focus on building machines with a toolbox for common sense,” he continued.
As for competitors, Marcus seems to think there aren’t many: “We don’t know anyone else trying to do this. Most often what happens nowadays is that when one wants to build a robot, one hacks together a mixture of ROS [Robot Operating System] and TensorFlow or PyTorch, tailored to a very specific problem. We don’t know of anyone building the kind of general-purpose cognitive tools we have in mind. There is no extant tool that can deliver the kind of complex, flexible cognition that we are focusing on.”