• robotics
• robot collision
• motion planning

In the last decade, the robotic research starts moving from the classical manipulators mounted inside a cage and strictly separated from the operators to smaller machines able to coexist and physically interact with humans. This trend is motivated on the short term by the reduction of the safety constraints imposed by the deployment of robotic manipulators in the workplace, making it more agile in term of following the needs of the market, while on the long term the research on this field aims to the spreading of robotics platform in daily life activities of peoples. A large part of the work in this direction is related to the generation of safe and trustworthy motions in robotic manipulators. One of the solutions proposed to this aim is to generate movements similar to the ones produced by humans, which can be predicted accepted by persons in an easier way. However, in unstructured time-variant environments is difficult to guarantee the avoidance of any collision during motion. For this reason, the robotic platform which works in this type od scenarios should be able to rapidly detect and react to unforeseen collisions and accomplish the desired task. A possible approach could be to add tactile sensors on the whole surface of the manipulators to get this information. However, this solution increases the complexity and the cost of the system. To avoid this problem, there are approaches in literature which exploits only proprioceptive information from the manipulator (i.e., angular positions and velocities) to estimate the contact with obstacles. In my thesis, starting from an already existing human-like motion planning algorithm, I proposed a framework able to rapidly detect and react without the usage of tactile sensing and, after that, plan a new trajectory exploiting the new information gathered from the contact with the environment. I tested this framework both in simulation and with a real manipulator proving the effectiveness of the proposed solution.