trevor-richardson Goto Github PK

collision_anticipation

This repository implements a convolutional recurrent neural network that learns the functional mapping between video data and the probability of future collisions. The labels used to train this deep learning algorithm are generated by a self-supervised collision-detection deep learning method.

collision_avoidant_reinforcement

This repository implements a deep reinforcement learning algorithm designed to utilize uncertainty in robotic model dynamics to learn to avoid future collisions.

conv_lstm_cell

Custom built ConvLSTM cell in Tensorflow and PyTorch

deep_dynamics

This repository implements a general data-driven framework for robotic-collision detection. Train a neural network to regress future states given the current state and action. Stochatsic forward passes are used at inference time in order to produce a belief distribution over the future next state of the robot. An exponentially smoothed norm is used between the set of regressed future states and the ground truth collected in the next time-step.

educational_material

Deep learning tutorials and education material for robotic scenarios -- Intel Sponsored

mcflow

Monte Carlo Flow Models for Data Imputation

misgan

MisGAN: Learning from Incomplete Data with GANs

rnn_zoo

This repository tests various recurrent neural network architectures on baseline datasets SeqMNIST and pMNIST.

spectral-lstm

Disentangle cell state in recurrent models to be a matrix of shape MxN and utilize SVD to extract features and their strengths

vasconomics_materials

vectorbt

Ultimate Python library for time series analysis and backtesting at speed and scale