Learning to Look Around: Intelligently Exploring Unseen Environments for Unknown Tasks

Abstract

It is common to implicitly assume access to intelligently captured inputs (e.g., photos from a human photographer), yet autonomously capturing good observations is itself a major challenge. We address the problem of learning to look around: if a visual agent has the ability to voluntarily acquire new views to observe its environment, how can it learn efficient exploratory behaviors to acquire informative observations? We propose a reinforcement learning solution, where the agent is rewarded for actions that reduce its uncertainty about the unobserved portions of its environment. Based on this principle, we develop a recurrent neural network-based approach to perform active completion of panoramic natural scenes and 3D object shapes. Crucially, the learned policies are not tied to any recognition task nor to the particular semantic content seen during training. As a result, 1) the learned "look around" behavior is relevant even for new tasks in unseen environments, and 2) training data acquisition involves no manual labeling. Through tests in diverse settings, we demonstrate that our approach learns useful generic policies that transfer to new unseen tasks and environments. Exploration episodes are shown at https://goo.gl/BgWX3W .

Emergence of exploratory look-around behaviors through active observation completion

“…As a core solution to these challenges, we present a reinforcement learning (RL) approach for active observation completion (23). See Figure 2.…”

“…wherex t (X, θ (i) ) denotes the reconstructed view at viewpoint θ (i) and time t, d denotes the pixelwise reconstruction MSE, and ∆ 0 denotes the offset to account for the unknown starting azimuth (23).…”

ShapeCodes: Self-supervised Feature Learning by Lifting Views to Viewgrids

“…Decoder: To learn a representation with this property, the output of the encoder is processed through another fully connected layer to increase its dimensionality The complete architecture, together with more detailed specifications, is visualized in Fig 2. Our convolutional encoder-decoder [43] neural network architecture is similar to [30,58,66,69]. As discussed above, however, the primary focus of our work is very different.…”

“…How then can it know the correct viewpoint coordinates for the viewgrid it must produce? It instead produces viewgrids aligned with the observed viewpoint at the azimuthal coordinate origin, similar to [30]. Azimuthal rotations of a given viewgrid all form an equivalence class.…”

Snap Angle Prediction for 360$$^{\circ }$$ Panoramas

“…The influential work of [25] learns a policy for visual attention in image classification. Active perception systems use RNNs and/or reinforcement learning to select places to look in a novel image [26,27], environment [28,29,30], or video [31,32,33,34] to detect certain objects or activities efficiently. Broadly construed, we share the general goal of efficiently converging on a desired target "view", but our problem domain is entirely different.…”