An AI enabled human online model for human-robot collaboration

Consumer markets today demonstrate an observable trend towards mass customisation and personalisation. Assembly processes are required to adapt, to meet the requirements of increased product complexity and constant variant updates. A concept to meet these challenges is a close collaboration between human workers and robots. Human-Robot Collaboration aims to combine the best of both worlds: the strength, precision, and endurance of robots with the sensory skills, problem solving, natural intuition and adaptability of humans. Moreover, this is also intended to overcome weakness associated with each party. On one hand, robots cope well with standardised tasks, yet require re-programming for new variants or tasks. On the other hand, human operators typically cope well with tasks that require a high level of flexibility and adaptability. Yet, monotonous and repetitive tasks can lead to high perceived workloads, fatigue, and stress, which result in a lowered job satisfaction and productivity. Thus, by combining the strengths of both heterogeneous parties, and enabling them to compensate each other’s weaknesses, synergetic or symbiotic effects are envisioned. To establish a symbiotic Human-Robot Collaboration, two main challenges were identified. Namely: effective task distribution/organisation and safety. In both cases, mutual awareness is essential, which becomes particularly challenging due the complementary differences of both parties. Ideally, the human and robot would “understand” each other. In this thesis, the prospect of equipping the robot with a model of the human is investigated. In particular, data-driven or so called online models of the human are chosen, which aim to capture and interpret biological signals as they occur in real-time. Thus, wearable sensors are being utilised. However, interpretation of these signals becomes greatly challenging, due to high levels of noise and subject specific characteristics within the data. To cope with these challenges, Machine Learning is utilised, which aims to establish adaptive models for each individual. In a first step, a generic human online model framework including a Machine Learning methodology was established. This framework was then applied in three scenarios, namely for detecting the human state (Perceived workload) since it affects task performance, predicting human movement intentions to better coordinate the robot’s behaviour, and finally for detecting the human operator noticing or anticipating potential emergencies. Overall, these online models of the human demonstrated a promising potential towards a safer and more effective teamwork between humans and robots.