Effective human–machine collaboration is a long standing problem within rehabilitation robotics. Machines that affect peer-to-peer interaction hold promise. Agents can serve as the basis for such machines. Belief–desire–intention (BDI) paradigm is used to achieve human-like intelligence. However, the BDI model does not emulate human-like qualities such as coordination or collaborative planning. Therein lies the motivation to extend the basic BDI agent model. cBDI—an extension to the BDI architecture supporting human–machine collaboration is presented. Two different applications for the cBDI agent are highlighted: (1) human–cBDI agent collaboration to achieve block stacking and (2) cBDI-based collaborative navigation for an intelligent wheelchair (IW). Under scenarios such that the block stacking task cannot be completed by either human or agent alone, we show how the cBDI is tuned to collaborate and accomplish the task. The cBDI-based collaborative navigation controller is for affecting assistance-as-required in an IW. The effectiveness of the cBDI-based collaborative architecture is evaluated within a ROS-USARSim framework.