Service Robot Adaptation to Users with Different Abilities

Service robots are expected to interact with users in a number of scenarios, from homes to offices and hospitals. Research in service robotics is constantly progressing towards generally more competent autonomous robots, able to perform a larger set of tasks, regardless of the people they interact with. People, however, have a wide range of different abilities, depending on age, health or medical conditions, which may require the robot’s behaviour to be tailored to each user. Humans quickly realize and adapt to such situations, and, for instance, would not ask an elderly person to lift a heavy weight, or a person on a wheel chair to take the stairs.

Previous research has considered the general problem of planning under different user’s preferences, requiring to model such preferences explicitly beforehand (which can include disabilities, as an elderly person could be seen as preferring not to lift heavy weights), including by asking questions to the user. However, such modelling may not always be possible or appropriate (questions about disabilities are particularly sensitive), and unknown users may require to interact with the system on short notice. Some interactions are short-lived (e.g., asking a robot for direction), while others are symbiotic and potentially years long (e.g., assisting people in a care home). Online adaptation of the robot to the particular needs of a user is therefore necessary to ensure safe human-robot collaboration, and trust in the robotic system, for a diverse and inclusive set of users.

In this project, we will create a framework for online adaptation through both planning and reinforcement learning, so that the system will be able to reason with symbolic models, which enables to verify and explain behaviours, and also adapt on the fly to the different abilities of the user it is interacting with. Short and long interactions pose different challenges, and we will study what aspects of vision, dialogue, and physical actions can be leveraged to adapt behaviours in a continual and life-long manner, while being respectful of the user’s conditions and privacy.

As service robotics deployment gains momentum (particularly within hospitals and care homes recently) it is the right time to ensure that the focus is not only on competent execution of specific tasks, but also on respectful and inclusive interaction with the people collaborating or being helped by the robot.

Piyush Khandelwal, Shiqi Zhang, Jivko Sinapov, Matteo Leonetti, Jesse Thomason, Fangkai Yang, Ilaria Gori, Maxwell Svetlik, Priyanka Khante, Vladimir Lifschitz, J. K. Aggarwal, Raymond Mooney, Peter Stone. BWIBots: A platform for bridging the gap between AI and human–robot interaction research. The International Journal of Robotics Research, 2017. https://doi.org/10.1177%2F0278364916688949

Matteo Leonetti, Luca Iocchi, Peter Stone. A synthesis of automated planning and reinforcement learning for efficient, robust decision-making. Artificial Intelligence, 2016. https://doi.org/10.1016/j.artint.2016.07.004

Canal, G., Alenyà, G., Torras, C. Adapting robot task planning to user preferences: an assistive shoe dressing example. Autonomous Robots,2019. https://doi.org/10.1007/s10514-018-9737-2

Project ID

STAI-CDT-2021-KCL-9