Peripersonal Space in the Humanoid Robot iCub

Abstract

Developing behaviours for interaction with objects close to the body is a primary goal for any organism to survive in the world. Being able to develop such behaviours will be an essential feature in autonomous humanoid robots in order to improve their integration into human environments. Adaptable spatial abilities will make robots safer and improve their social skills, human-robot and robot-robot collaboration abilities.

This work investigated how a humanoid robot can explore and create action-based representations of its peripersonal space, the region immediately surrounding the body where reaching is possible without location displacement. It presents three empirical studies based on peripersonal space findings from psychology, neuroscience and robotics. The experiments used a visual perception system based on active-vision and biologically inspired neural networks.

The first study investigated the contribution of binocular vision in a reaching task. Results indicated the signal from vergence is a useful embodied depth estimation cue in the peripersonal space in humanoid robots. The second study explored the influence of morphology and postural experience on confidence levels in reaching assessment. Results showed that a decrease of confidence when assessing targets located farther from the body, possibly in accordance to errors in depth estimation from vergence for longer distances. Additionally, it was found that a proprioceptive arm-length signal extends the robot’s peripersonal space. The last experiment modelled development of the reaching skill by implementing motor synergies that progressively unlock degrees of freedom in the arm. The model was advantageous when compared to one that included no developmental stages.

The contribution to knowledge of this work is extending the research on biologically-inspired methods for building robots, presenting new ways to further investigate the robotic properties involved in the dynamical adaptation to body and sensing characteristics, vision-based action, morphology and confidence levels in reaching assessment.