Simulating human motions in industrial environments is costly, manual effort. Available solutions that automate modeling suffer from lacking naturalness. Data driven motion synthesis may solve this issue. However, it requires a large number of previously recorded motions as input.
This work investigates experimental effort for covering motion variability of picking actions observed on an actual automotive assembly shop floor. The gathering of the necessary data at the shop floor with feasible effort is depicted. A set of 17 motion styles is identified and analyzed for frequency of occurrences at an exemplary assembly station at an automotive OEM. From this analysis, an estimate for the lower bound of experimental effort in terms of required training data is derived. Considering an existing data driven human motion simulation approach, possibilities to minimize the number of experiments are discussed.