Published on Sun Jul 12 2020

OtoWorld: Towards Learning to Separate by Learning to Move

Omkar Ranadive, Grant Gasser, David Terpay, Prem Seetharaman

OtoWorld is an interactive environment in which agents must learn to listen in order to solve navigational tasks. To solve one episode of OtoWorld, an agent must move towards each sounding source in the auditory scene and "turn it off" The sources are placed randomly within the room.

0
0
0
Abstract

We present OtoWorld, an interactive environment in which agents must learn to listen in order to solve navigational tasks. The purpose of OtoWorld is to facilitate reinforcement learning research in computer audition, where agents must learn to listen to the world around them to navigate. OtoWorld is built on three open source libraries: OpenAI Gym for environment and agent interaction, PyRoomAcoustics for ray-tracing and acoustics simulation, and nussl for training deep computer audition models. OtoWorld is the audio analogue of GridWorld, a simple navigation game. OtoWorld can be easily extended to more complex environments and games. To solve one episode of OtoWorld, an agent must move towards each sounding source in the auditory scene and "turn it off". The agent receives no other input than the current sound of the room. The sources are placed randomly within the room and can vary in number. The agent receives a reward for turning off a source. We present preliminary results on the ability of agents to win at OtoWorld. OtoWorld is open-source and available.

Mon Jul 05 2021
Machine Learning
Agents that Listen: High-Throughput Reinforcement Learning with Multiple Sensory Systems
New version of VizDoom simulator provides raw audio observations. We study the performance of different model architectures in a series of tasks. We train our agent to play the full game of Doom and find that it can consistently defeat a traditional vision-based adversary.
3
16
51
Mon Apr 20 2020
Neural Networks
Learning as Reinforcement: Applying Principles of Neuroscience for More General Reinforcement Learning Agents
The Neurons-in-a-Box architecture can learn in a wholly generalizable manner. It demonstrates an efficient way to build representations without explicitly optimizing over a set of actions.
0
0
0
Wed Feb 21 2018
Artificial Intelligence
Learning to Play with Intrinsically-Motivated Self-Aware Agents
Infants are experts at playing, with an amazing ability to generate novel structured behaviors. We seek to mathematically formalize these abilities using a network that implements curiosity-driven intrinsic motivation. We demonstrate that this policy causes the agent to explore novel and informative interactions with its environment.
0
0
0
Wed Nov 16 2016
Neural Networks
Reinforcement Learning with Unsupervised Auxiliary Tasks
Deep reinforcement learning agents have achieved state-of-the-art results by maximising cumulative reward. However, environments contain a much wider variety of possible training signals. We introduce an agent that maximises many other pseudo-reward functions simultaneously by reinforcement learning.
0
0
0
Wed Jul 25 2018
Artificial Intelligence
Attend Before you Act: Leveraging human visual attention for continual learning
Humans selectively pay attention to certain parts of an image when playing a game. This information is used to build a representation of the scene.
0
0
0
Mon Oct 15 2018
Machine Learning
Deep Reinforcement Learning
We discuss six core elements, six important Mechanisms, and twelve applications. We discuss RL applications, including games, natural language processing (NLP), computer vision, finance, business management, healthcare, education, energy, transportation, computer systems, science, engineering, and art.
2
2
5