Abstract: The present disclosure provides an adaptive meta-learning technique for determining robotic action. Conventional methods are focusing on task-relevant aspects of the input observations and fails to provide adaptive learning. Initially, a plurality of images pertaining to a visual demonstration for a robot are received by the system. Further, a plurality of vector embeddings are computed based on the plurality of images using an attentive embedding network. The attentive embedding network includes a first Convolutional Neural Network (CNN), a fully connected layer and a plurality of spatial attention modules. Finally, a control action is computed based on the plurality of vector embeddings, an image from the plurality of images, a robot joint state vector and robot joint velocity vector using a control network. The control network comprises a second CNN and a plurality of fully connected layers. The control network is connected to the attentive embedding using multiplicative spatial skip connections.

Abstract: Depth estimation of images using deep learning methods is a wide range of application in Augmented Reality, 3D graphics and robotics. Conventional methods are supervised, which requires explicit ground truth depth information for training and the conventional unsupervised methods fails to provide a generalized solution. The present disclosure estimates accurate depth information and confidence map of a given monocular image in an unsupervised manner. A depth Neural Network receives a monocular image and predicts per-pixel depth map and a confidence map. The depth NN utilizes a negative exponential of photometric loss as ground truth information. The predicted confidence-map is further used to estimate per-pixel uncertainty map. The pose NN predicts a plurality of pose vectors between a plurality of the consecutive monocular images. Finally, the Bayesian inference module is computes the fused depth information and the fused uncertainty map.