Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Data-driven architecture to encode information in the kinematics of robots and artificial avatars

Francesco De Lellis, Marco Coraggio, Nathan C. Foster, Riccardo Villa, Cristina Becchio, Mario di Bernardo

Abstract

We present a data-driven control architecture for modifying the kinematics of robots and artificial avatars to encode specific information such as the presence or not of an emotion in the movements of an avatar or robot driven by a human operator. We validate our approach on an experimental dataset obtained during the reach-to-grasp phase of a pick-and-place task.

Data-driven architecture to encode information in the kinematics of robots and artificial avatars

Abstract

We present a data-driven control architecture for modifying the kinematics of robots and artificial avatars to encode specific information such as the presence or not of an emotion in the movements of an avatar or robot driven by a human operator. We validate our approach on an experimental dataset obtained during the reach-to-grasp phase of a pick-and-place task.
Paper Structure (14 sections, 1 theorem, 18 equations, 6 figures, 1 algorithm)

This paper contains 14 sections, 1 theorem, 18 equations, 6 figures, 1 algorithm.

Table of Contents

  1. Introduction
  2. State of the art
  3. Control of body movement to express emotion
  4. Preliminaries
  5. Problem statement
  6. A data driven solution framework
  7. Offline solution
  8. Online computation
  9. Enforcing initial and terminal conditions
  10. Heuristic solution to the problem of reaching an object
  11. Validation
  12. Training of the approximate encoding function $\hat{\varepsilon}$
  13. Computation of $\tilde{w}$ and enforcement of the terminal conditions
  14. Validation of the control strategy

Key Result

Lemma 1

Assume that the encoding function $\varepsilon$ is approximated by some $\hat{\varepsilon}$ such that Then, when $v_\mathrm{a}$ is computed from eq:altered_motion, the solution $c^*$ to Problem eq:problem_statement_implemented yields a $v_\mathrm{a}$ that is optimal for Problem eq:problem_statement.

Figures (6)

  • Figure 1: Block scheme of the proposed strategy to solve Problem \ref{['eq:problem_statement']}.
  • Figure 2: Example of the trajectories, projections, restrictions, and expansions described in Sec. \ref{['sec:online_computation']}.
  • Figure 3: Relevant quantities concerning the training of the approximate encoding function (see Sec. \ref{['sec:training_approximate_encoding_function']}). Results are computed as an average of the 5 sessions used in cross validation.
  • Figure 4: Moving average of 100-sample cumulative discounted rewards per episode. Red line: threshold value $\sigma = 10000$, defined in de2023guaranteeing. The agent surpasses the threshold after 800 episodes, indicating successful constraint enforcement \ref{['eq:constr_finite_time_convergence']} on training examples in $\mathcal{D}_{\neg e_\mathrm{des}}$.
  • Figure 5: Values of the restricted solution function $\tilde{w}(1, v_\mathrm{h}, v_\mathrm{r})$ for $v_\mathrm{h} \in \mathcal{D}_{\neg e_\mathrm{des}}$ and $v_\mathrm{r} \in \mathcal{D}_{e_\mathrm{des}}$ (see Sec. \ref{['sec:heuristic_solution']}).
  • ...and 1 more figures

Theorems & Definitions (2)

  • Lemma 1
  • proof