Robots and voice commands
Robots are already helping elderly or disabled people with household tasks. Yet, to truly benefit from such technologies, users should be able to control domestic robots using voice commands. The EU-funded ROBOTASK (Action words learning in a humanoid robot by discovering tool affordances via statistical inference) project developed a language model for robots. The model enabled robots to process sentences describing actions. Additionally, the team’s human-robot interaction studies investigated the feasibility of a robot so-endowed participating in a domestic task as a partner. The team considered whether the robot could intervene in an action performed by a human. Researchers proposed a model that grounds language into objects and motor sequences (called affordances). Modelling affordances permitted study of the effects of trapping objects, and of pulling objects using a tool. Using the model, robots can estimate the effects of actions upon an object. Following from the idea that concepts are encoded in language- and sensorimotor-based representations, researchers proposed an embodied statistical language model incorporating both kinds of knowledge. The model enabled the grounding of symbols in sensorimotor knowledge, and a semantic layer for reasoning about the symbols. The ROBOTASK team also proposed an affordance-based planner, to be implemented through hierarchical task networks. The planner would enable the robot to develop a strategy for implementing its part in a human-robot joint task. The robot would have developed a simple anticipation of a human’s needs, plus some decision-making capacity concerning when to intervene in the human’s actions. As a result, robots should be able to assist their human partners in a shared goal. The robot would be able to reason about both sides of an action, and offer contextually appropriate help. Aside from domestic assistance, ROBOTASK’s work may find application in industrial semi-automation. A new generation of robots would collaborate with humans on tasks that cannot be automated, combining the best of human cognition with robot skills.
Keywords
Robots, voice commands, language model, ROBOTASK, affordances, human-robot
