Boston Dynamics SPOT and Microsoft Teams to automate inspections

Inspection is a fundamental task in industrial plants. Typically, there are mainly visual and measuring inspections. The first type aims just at checking the physical conditions of the facilities, while the second involves some kind of measurement or tool reading. Whatever the case, repetitiveness is a critical aspect of inspections, that could lead to errors due to lack of attention if performed by human operators. Additionally, these are time consumption and potentially hazardous tasks. In this context, the employment of mobile robots for inspection is constantly increasing in modern companies, but in many situations the human-robot cooperation is still needed, since a complete automation is not yet possible, mainly due to the unpredictable changes in the working area. Main types of mobile robots involved for inspections are drones, wheeled and legged, all with their advantages and drawbacks, usually equipped with camera and sensors that together with computer vision and AI give the robots the possibility to see and understand the surroundings. A typical use case scenario for plant inspections could be an operator telling robot A to start with the inspection, then, if needed, the robot itself asks for the intervention of a robot B equipped with specific payload to intervene in a particular and critic situation. Considering this scenario and in the context of enhancing the quality and the level of automation of inspections, the goals of this project are: to make easier the human-robot interaction through the possibility of send commands to a robot exploiting user-friendly tools like a text chat and without a specific training for the user, giving the operator the possibility to talk to the robot in an pseudo-natural way. Automatize a specific task like the manipulation of a knob, involving the automatic generation a custom dataset for neural network training aimed at specific target object recognition. Finally, enabling the human-independent cooperation between robots, potentially orchestrating a robot fleet. The result obtained from this project, which is a collaboration between Politecnico di Torino and Sprint Reply, was the possibility to interact with Spot from Boston Dynamics through a Microsoft Teams chat, asking to perform an inspection, and automatically triggering the intervention of another Spot equipped with robotic arm to close a knob if water coming from the relative pipe is detected. Additionally, a complete pipeline for automatic custom dataset generation starting from few sample images was developed, with the aim of improving the recognition of specific objects with YOLO neural network