A project synopsis on

A Generalised Deep Reinforcement Algorithm Developed using Virtual Robotic Computer

simulations & various world markets

SYNOPSIS

The successes of deep reinforcement learning in recent years have led many researchers across
various engineering fields to develop methods to control robots and markets using variations of
DRL. RL is an area of artificial intelligence concerned with agents ought to take actions in an
environment so as to maximize some notion of cumulative reward. In other words, It’s a technique
to automate the process of designing,sensing, planning and controlling just by letting the robot learn
by themselves autonomously so as to achieve a generalised algorithm for everything . But
developing individual physical mechanical systems to train the algortihm for each task, increases
the cost of the project rapidly. Due to this problem , In this project we have chosen virtual robotic
stimulations and various world stocks and commodities market ( such as Al,Cu,Pb,Sn) to test our
hypothesis. Traditionally, the go-to approach when desiring extremely low sample complexity in RL
is model-based RL. In model-based RL, instead of learning a policy only based on rewards obtained
by interaction with the environment, the agent tries to learn a model of the environment and use it to
plan and improve its policy, thus dramatically reducing the number of interactions it needs with the
environment. However, this approach often results in lower asymptotic performance compared to
model-free algorithms, and sometimes suffers from catastrophic failures due to errors in the learned
model. DRL algorithm learns both a value network (which predicts the winner) and a policy
network (which selects actions) through virtual simulations . We are using DRL methods for
continuous control problems such as robotic manipulation and locomotion.