Autonomy: from Shakey to the Future
The Shakey robot, built at SRI International in the late 1960’s and early 1970’s, was the world’s first mobile robot to perceive its environment, figure out how to execute goals given by the user, and then perform actions. Artificial intelligence (AI) researchers today are still working on these problems, in ever more complex environments and for ever more complex tasks.
We have seen spectacular progress in fundamental areas like perception, navigation and mapping, mobility, and manipulation. Indeed, autonomy is booming, from driverless cars to robotic appliances and the beginning of effective humanoid robots. But for these autonomous systems to be successfully integrated into our everyday lives, we must address three critical challenges:
- Interaction – Advances in human-machine interaction have captured the public imagination in systems like Siri, Alexa, etc. But this is only the beginning of what we need if we are to really tell a robot (or any computer system) what to do. How do we explore what the system does or doesn’t know? How does the system ask questions and request guidance? This is the essence of human conversation -- verbal and non-verbal -- and we are still working on how to get computers to play their part.
- Collaboration – Collaboration goes beyond communication. Shakey collaborated with humans in only the most basic sense (taking orders). Real collaboration requires shared understanding of capabilities and plans. We need robots that can collaborate with people and other robots to perform tasks, and be companionable. For example, a home robot for an elderly person needs to be sensitive to that person's interests, capabilities, and current state.
- Trust – How can we trust the complicated systems we are building? Autonomous vehicles are making the headlines, but the issue is there for any complex system. It's even more poignant for robots, because we will be living with them constantly, and because we have an irresistible urge to anthropomorphize them (thus raising the ante on trust). Moving toward trust requires a multifaceted approach: we must build systems in ways that make them more transparent and predictable; we must provide mechanisms for mapping between the system’s understanding and our understanding (e.g., the system must be able to explain its intents and actions); and we must protect the system from external attacks that can undermine our trust (cyber and physical security).
SRI has significant research in these three interrelated areas. We are designing and delivering interactive systems with greater conversational ability, based not only on speech and text, but also on perception of gestures and other non-verbal communication. One of our thrust areas is increasing system awareness of and adaptation to human states like fatigue, engagement, confusion, and emotion – vital to natural interaction with robots and other computer systems.
We are designing collaboration environments for human-robot and robot-robot teams. We are building systems that enable humans to give advice to autonomous systems to guide their choices and actions. One aspect of collaboration is the ability of all participants to initiate interaction to ask for help and guidance, which requires systems to have some self-understanding of their capabilities and limitations.
Finally, we are working on making systems more trustable. We are building on decades of research on how to assure correct system behavior and provide fault tolerance. We are developing techniques to make system policies and reasoning more transparent, including systems that can explain their potential decisions. As part of our cyber security research, we are emphasizing protection of cyber physical systems, including autonomous systems and the Internet of Things.
One application of special interest for this research is support for the elderly population. Researchers at SRI are focusing on projects aimed at providing interactive companionship, privacy-preserving health monitoring, and aids for daily living.
Shakey demonstrated 50 years ago that real autonomy was feasible. As we race toward practical autonomy, we need to accelerate the research required to fulfill the vision of making autonomous systems our most important collaborators.