User-Specific Adaptation of Collaborative Robot Motion for Improved Ergonomics – ROOIBOS

The objective is to propose a methodology for planning cobot actions in order to minimize long-term physical fatigue for the user, while taking into account uncertainties regarding human postural reactions and estimated fatigue levels. This method is based on three elements:

- the formal POMDP (Partially Observable Markov Decision Process) framework, which enables the planning of long-term strategies in uncertain conditions, with partial observability;

- a digital human simulator based on a physics engine that calculates the joint torques of the human during a movement in response to a robot action, which is necessary to anticipate the cost of the various possible actions. This simulator is based on a linear quadratic programming (LQP) control technique that generates full-body movements of the simulated human (as well as the associated internal forces) from a high-level description of the tasks to be performed (e.g., maintaining balance, following a trajectory with one hand). This technique also makes it possible to simulate different ways of performing the movement (i.e., different postural strategies) by adjusting the elements included in the objective function (e.g., penalizing deviation of the back from the vertical, penalizing movements of certain joints relative to others);

- A fatigue model from the literature that estimates physical fatigue based on the joint torques exerted by humans. Combining this model with the simulator mentioned above makes it possible to estimate the physical fatigue, at the level of the various human joints, caused by each of the possible postural reactions of humans to each possible action (in our case, the placement of an object) of the cobot. This predicted fatigue is then used to solve the POMDP problem and calculate the optimal policy for the cobot.

The proposed approach was validated in simulation, using a sample task. We compared its performance, in terms of reducing physical fatigue accumulated during a repetitive task, with other possible policies for robot behavior: a fixed policy (always performing the same action), a cyclical policy (alternating between different possible actions), a random policy, and a “short-sighted” policy (optimizing fatigue over the horizon of a cycle, i.e., only in the short term) . We showed that the proposed approach based on long-term optimization significantly reduces accumulated fatigue compared to all other policies, especially when the task is highly repetitive (large number of cycles). We have demonstrated this for different situations of uncertainty: with and without uncertainty about the initial state of human fatigue, and in the case of a choice of postural response by the human that is either deterministic (without uncertainty) or stochastic (with uncertainty).

Experimental validation of this approach is currently underway.

This project opens up numerous research opportunities on the customization of robotic assistance for professional movements and the consideration of human movement variability in cobotics.

Collaborative robots have the potential to reduce work-related musculoskeletal disorders not only by decreasing the workers' physical load, but also by modifying and improving their postures. Imposing a sudden modification of one's movement can however be detrimental to the acceptance and efficacy of the human-robot collaboration. In ROOIBOS, we will develop a framework to plan user-specific trajectories for collaborative robots, to gradually optimize the efficacy of the collaboration and the long-term occupational health of the user. We will use machine learning and probabilistic methods to perform user-specific prediction of whole-body movements. We will define dedicated metrics to evaluate the movement ergonomic performance and intuitiveness. We will integrate those elements in a digital human simulation to plan a progressive adaptation of the robot motion accounting for the user's motor preferences. We will then use probabilistic decision-making to adapt the plan on-line to the user's motor adaptation capabilities. This will enable a smooth deployment of collaborative robots at work.