we introduce an interactive method to integrate automatic behavior and teleoperation. In various environments, automatic behavior programmed in advance might not work. Humanoids need support by human using tele-operation. On the other hand, operability would be better with support of humanoid in tele-operation. We propose stochastic method to resolve when and how humanoids give up and get over initiative. We actually implemented the method on a humanoid; HRP-2W.

As a example for the shift from autonomy to tele-operation, we proposed that HRP-2W should ask the position of objects when he failed to recognize it. On this occasion, HRP-2W should tell human what kind of information he needs for appropriate support.

As for the shift from tele-operation to autonomy, we demonstrated that HRP-2W should take over the initiative when operator's command seemed to be wrong. HRP-2W computes the certainty factor from operator's commands of the past. This system uses Inamura's "PEXIS", which utilizes Bayesian network.