Current techniques of human-robot object transfer usually consist in following a trajectory completely defined before the motion starts, limiting thus the capacity to adapt the motion plan to the uncontrolled human behavior. Furthermore, the robotic arm motions usually do not rely on characteristics of the human motion dynamics, while the object transfer itself is generally realized without using sensing information as the human does with his hand.


CogLaboration proposes to address this challenge by: 

 

  • Studing the characteristics of successful human-human object exchange in realistic task settings and conditions, including typical variations and unplanned and unanticipated situations.
  • Developing a hierarchical control architecture based on concepts from cognitive neurosciences and exploiting the key-characteristics of the human-human exchange
  • Building a vision-driven robotic system comprising a lightweight arm and a hand with tactile sensors