Stochastic optimal feedforward-feedback control determines timing and variability of arm movements with or without vision

Abstract

Human movements with or without vision exhibit timing (i.e. speed and duration) and variability characteristics which are not well captured by existing computational models. Here, we introduce a stochastic optimal feedforward-feedback control (SFFC) model that can predict the nominal timing and trial-by-trial variability of self-paced arm reaching movements carried out with or without online visual feedback of the hand. In SFFC, movement timing results from the minimization of the intrinsic factors of effort and variance due to constant and signal-dependent motor noise, and movement variability depends on the integration of visual feedback. Reaching arm movements data are used to examine the effect of online vision on movement timing and variability, and test the model. This modelling suggests that the central nervous system predicts the effects of sensorimotor noise to generate an optimal feedforward motor command, and triggers optimal feedback corrections to task-related errors based on the available limb state estimate.

Adapting the visuo-haptic perception through muscle coactivation

“…Importantly, while the subjects could have coactivated their wrist indiscriminately to yield superior performance through haptic guidance to the target, they only increased coactivation to maintain the same performance level independently of the visual noise level. This points to a trade-off between task performance and effort, a common metric used to explain human motor behaviors, from reaching to interaction tasks 16 – 18 .…”

Optimal Feedback Control for Modeling Human-Computer Interaction

“…Peak velocity and acceleration are also very close to the values of the respective user trajectory, albeit the maximum acceleration is higher and the timing of the minimum acceleration does not match exactly. However, it is important to note that the duration of the surge phase was not explicitly built into the LQR model 9 , but emerges naturally from the interplay of the optimal parameters 𝜔 𝑣 , 𝜔 𝑓 , and 𝜔 𝑟 . In contrast, in the MinJerk model, the duration of the surge phase needs to be either known in advance or determined using the parameter fitting process.…”

“…The infinite-horizon LQR/LQG formulation [102] is less suitable for many HCI tasks, as it does not allow to take into account multiple, time-dependent objectives during optimization, which, e.g., is inevitable for via-point tasks that need to be reached in a given order, or moving targets. However, the class of optimal control models of Human-Computer Interaction, as discussed in Section 3, is much larger and consists of a variety of modeling approaches and solution methods, including Direct and Indirect Collocation [11], Model-Free and Model-Based Reinforcement Learning [52,119], (Semi-)Supervised Learning [92,106], Model-Predictive Control [20], and mixtures of these [9,12,76], each of which has its own requirements on the problem, advantages, and disadvantages. While some HCI tasks might be too complex to solve using the linear methods presented in this paper, the general OFC framework offers exciting opportunities to model, simulate, explore, and eventually improve the interaction between humans and computers, using a mathematically profound description.…”

“…We have tried to include the system dynamics parameters 𝑚, 𝜏 1 , and 𝜏 2 in the parameter fitting process. We found that this introduces more redundancy within the set of possible parameter combinations, resulting in additional (close-to-)optimal parameter vectors that are often unrealistic (e.g., an extremely large mass or very large time constants), while the fit to observed user behavior did not improve considerably 9. In theory, it would be possible to include prior knowledge about submovements and thus induce time-dependent behavior by choosing the time-dependent cost matrices 𝑄 𝑛 and 𝑅 𝑛 appropriately.…”

An approximate stochastic optimal control framework to simulate nonlinear neuro-musculoskeletal models in the presence of noise

“…Berret et al have applied this approach to simulations of reaching [ 30 ]. However, they only considered feedforward control [ 30 ] or added feedback control to the pre-computed optimal feedforward control in a post-processing step using the LQG framework [ 31 ]. This does not seem to yield an optimal solution as during the optimization of the feedforward control policy, the presence of feedback is neglected.…”