Sanctuary Cognitive Systems Raises $58.5M to Embed ‘Human-Like Intelligence’ in Robot Workers
Robotic AI startup Sanctuary Cognitive Systems has closed a $58.5 million Series A funding round of strategic investment from Bell, Verizon Ventures, and a handful of other corporate giants. Sanctuary’s research on ‘human-like intelligence’ is aimed at creating general-purpose robots that can interact with and learn from humans to work in different industries. The company pitches the AI as a potential aid for jobs that are too hard or dangerous, even as a way to make up for labor shortages, though wide deployment is not likely in the near future.
Humans are capable of learning a lot of widely varying ideas and activities, while robots are almost always built for a singular purpose or a narrow range of tasks. Sanctuary wants to produce an AI within a robot that is more like a human in terms of general-purpose roles and flexible thinking, one capable of understanding and learning how to do things like humans. That’s why the robots are humanoid, to match human thinking to a human form. That way they will be able to mimic and learn how humans do things, even if it’s in environments and circumstances that pose difficulties and dangers to humans like space or the ocean. That includes learning through explanation and demonstration using conversational AI tools.
“With unfilled vacancies, workplace safety considerations, increasing employee turnover, worldwide aging populations, and declining workplace participation, one thing is clear: many labor-related challenges are outside the scope of current specialized AI and robotics technology,” Sanctuary CEO Geordie Rose said. “We are addressing a systemic problem across the global economy. I am excited about the group of industry partners and investors we assembled. With interest from customers representing a dozen different industry verticals, we are working hard to make work safer, more accessible, and ultimately more productive.”
Robots that can adapt their behavior to circumstances and learn from instructions is an enormous goal, but one that many are experimenting with bringing to life. Amazon, in particular, has been keen to explore the potential with Alexa. That’s where features like Teachable AI, which lets users directly instruct Alexa about their preferences, and latent goal inference for extrapolating customer desires come from. And it’s the explicit goal of the SimBot Alexa Prize. The SimBot contest focuses on coming up with machine-learning models using natural language understanding that can teach a robot how to interact with humans and complete a task. The idea is that the AI will be able to learn from humans and from its own efforts to navigate and understand the space around it so it can manipulate objects and complete its assigned tasks in both virtual and real space.
The difficulties of humanoid robots have stymied many companies already. Softbank semi-retired the humanoid robot named Pepper recently partly for that reason. Elon Musk’s plans for a humanoid Tesla Bot, one he wants to see prototyped before the end of the summer, faces the same issues. Sanctuary’s AI will need to overcome those obstacles to truly become a part of any workforce.