Task-Oriented Probabilistic Active Vision

Abstract

In this work, an explicitly task-oriented approach to the active vision problem is presented. The system tries to reduce the most relevant components of the uncertainty in the world model, for the task the robot is currently performing. It is task oriented in the sense that it explicitly considers a task-specific value function. As test-bed for the presented active vision approach, we selected a robot soccer attention problem: goal-covering by a goalie player. The proposed system is compared with information-based approaches. Experimental results show that it surpasses them in the tested application. We conclude that, when the goal is not the uncertainty reduction itself, the minimization of the belief entropy is not a useful optimality criterion, and that for such cases, task-oriented optimality criteria are better suited.

Active sensing for continuous state and action spaces via task-action entropy minimization

“…So, the ultimate goal is not to obtain information but to perform the task well. Finally, the proposed method is different from the work in [12] and [13] because no training or value function is required. Instead, the proposed method only assumes the availability of a policy that maps a state into a task-related action.…”

“…The robot then performs the best action according the learned action-value function. Guerrero [13] presents an active-sensing method that selects sensing actions that maximize the expected value, where, a value function over the states is calculated via value iteration.…”

Task-Oriented Active Sensing via Action Entropy Minimization

“…The robot then performs the best action according to the learned actionvalue function. Guerrero et al present two task-oriented active sensing methods for soccer robots [15]. The difference between the two methods is the objective function.…”

Evolutionary active vision system: from 2D to 3D

“…Minut and Mahadevan 2001). However, there are also others that are instances of a more general approach such as the probabilistic approach (see Vidal-Calleja et al 2010;Guerrero et al 2010;Dame and Marchand 2013;Davison 2005) and adaptive approach (see Mirolli et al 2010;Kato and Floreano 2001;Croon 2008;Tuci 2014).…”

“…For instance, there are models for detecting edges (Kass, Witkin, & Terzopoulos, 1988), for controlling the gaze of a simulated fish (Terzopoulos & Rabie, 1995) and for detecting an object in a visual scene (Minut & Mahadevan, 2001). However, there are also others that are instances of a more general approach such as the probabilistic approach (Dame & Marchand, 2013; Davison, 2005; Guerrero, Ruiz-Del-Solar, Romero, & Angulo, 2010; Vidal-Calleja, Sanfeliu, & Andrade-Cetto, 2010) and adaptive approach (Croon, 2008; Kato & Floreano, 2001; Mirolli et al, 2010; Tuci, 2014).…”