Gesteme-free context-aware adaptation of robot behavior in human-robot cooperation.

BACKGROUND: Cooperative robotics is receiving greater acceptance because the typical advantages provided by manipulators are combined with an intuitive usage. In particular, hands-on robotics may benefit from the adaptation of the assistant behavior with respect to the activity currently performed by the user. A fast and reliable classification of human activities is required, as well as strategies to smoothly modify the control of the manipulator. In this scenario, gesteme-based motion classification is inadequate because it needs the observation of a wide signal percentage and the definition of a rich vocabulary. OBJECTIVE: In this work, a system able to recognize the user's current activity without a vocabulary of gestemes, and to accordingly adapt the manipulator's dynamic behavior is presented. METHODS AND MATERIAL: An underlying stochastic model fits variations in the user's guidance forces and the resulting trajectories of the manipulator's end-effector with a set of Gaussian distribution. The high-level switching between these distributions is captured with hidden Markov models. The dynamic of the KUKA light-weight robot, a torque-controlled manipulator, is modified with respect to the classified activity using sigmoidal-shaped functions. The presented system is validated over a pool of 12 näive users in a scenario that addresses surgical targeting tasks on soft tissue. The robot's assistance is adapted in order to obtain a stiff behavior during activities that require critical accuracy constraint, and higher compliance during wide movements. Both the ability to provide the correct classification at each moment (sample accuracy) and the capability of correctly identify the correct sequence of activity (sequence accuracy) were evaluated. RESULTS: The proposed classifier is fast and accurate in all the experiments conducted (80% sample accuracy after the observation of ∼450ms of signal). Moreover, the ability of recognize the correct sequence of activities, without unwanted transitions is guaranteed (sequence accuracy ∼90% when computed far away from user desired transitions). Finally, the proposed activity-based adaptation of the robot's dynamic does not lead to a not smooth behavior (high smoothness, i.e. normalized jerk score <0.01). CONCLUSION: The provided system is able to dynamic assist the operator during cooperation in the presented scenario.

Redundancy optimization strategy for hands-on robotic surgery.

During hands-on cooperative surgery, the use of a redundant robot allows to address encumbrance issues in the Operating Room (OR), which can occur due to the presence of large medical instrumentation, such as the surgical microscope. This work presents a new Null Space Optimization (NSO) strategy to constraint the position of the manipulator's elbow within predefined range of motions, according to the spatial requirements of the specific procedure, also taking into account the physical joint limits of the robotic assistant. The proposed strategy was applied to the 7 degrees of freedom (dof) lightweight robot LWR4+. The performance of the NSO was compared to two state-of-the-art null space optimization strategies, i.e. damped posture and fixed optimal posture, over a pool of three non-expert users in both strict (20deg) and negligible (100deg) angular encumbrance limitations. The NSO strategy was proved versatile in providing wide elbow mobility together with safe distance from relevant continuity null space boundaries, guaranteeing smooth guidance trajectories. Future works would be performed in order to evaluate the potential feasibility of NSO in a real surgical scenario.

Recognition of user's activity for adaptive cooperative assistance in robotic surgery.

During hands-on robotic surgery it is advisable to know how and when to provide the surgeon with different assistance levels with respect to the current performed activity. Gesteme-based on-line classification requires the definition of a complete set of primitives and the observation of large signal percentage. In this work an on-line, gesteme-free activity recognition method is addressed. The algorithm models the guidance forces and the resulting trajectory of the manipulator with 26 low-level components of a Gaussian Mixture Model (GMM). Temporal switching among the components is modeled with a Hidden Markov Model (HMM). Tests are performed in a simplified scenario over a pool of 5 non-surgeon users. Classification accuracy resulted higher than 89% after the observation of a 300 ms-long signal. Future work will address the use of the current detected activity to on-line trigger different strategies to control the manipulator and adapt the level of assistance.