Sample-efficient reinforcement learning for quadcopter flight control

The combination of reinforcement learning and deep neural networks has the potential to train intelligent autonomous agents on high dimensional sensory inputs, with applications in flight control. However, the amount of samples needed by these methods is often too large to use real-world interaction. In this work, mirror-descent guided policy...

Reinforcement Learning for Helicopter Flight Control

Large-scale helicopters have unique characteristics of maneuverability and low-speed performance compared to fixed-wing aircraft. They can take off and land vertically, hover in place for extended periods of time, and move in all six directions, making them occupy important niches in both military and civil aviation. However, these advantages...

Safe Reinforcement Learning in Flight Control: Introduction to Safe Incremental Dual Heuristic Programming

Online continuous reinforcement learning has shown promising result in flight control achieving near optimal control within seconds and the capability to adapt to sudden changes in the environment. However no guarantees about safety can be given, needed for use in general aviation. Furthermore performance is often dependent on the precise tuning...

Safe Reinforcement Learning Applications

Reinforcement Learning (RL) focuses on maximizing the returns (discounted rewards) throughout the episodes, one of the main challenges when using it is that it is inadequate for safety-critical tasks due to the possibility of transitioning into critical states while exploring. Safe Reinforcement Learning (SafeRL) is a subset of RL that focuses...

Longitudinal Flight Control by Reinforcement Learning: Online Adaptive Critic Design Approach to Altitude Control

Reinforcement learning is used as a type of adaptive flight control. Adaptive Critic Design (ACD) is a popular approach for online reinforcement learning control due to its explicit generalization of the policy evaluation and the policy improvement elements. A variant of ACD, Incremental Dual Heuristic Programming (IDHP) has previously been...

Hierarchical Reinforcement Learning for Model-Free Flight Control: A sample efficient tabular approach using Q(lambda)-learning and options in a traditional flight control structure

Reinforcement learning (RL) is a model-free adaptive approach to learn a non-linear control law for flight control. However, for flat-RL (FRL) the size of the search space grows exponentially with the number of states, resulting in low sample efficiency. This research aims to improve the efficiency with Hierarchical Reinforcement Learning (HRL)....

Obstacle avoidance for quadrotors using reinforcement learning and obstacle-airflow interactions

This research investigates and proposes a new method for obstacle detection and avoidance on quadrotors. One that does not require the addition of any sensors, but relies solely on measurements from the accelerometer and rotor controllers. The detection of obstacles is based on the principle that the airflow around a quadrotor changes when the...

Flexible Curriculum Learning for Deep Reinforcement Learning Agents

Reinforcement Learning (RL) is a learning paradigm that learns by interacting with the environment. In practice, a RL agent needs to perform many actions to sample rewards and state transitions from their environments. Recent advances in using deep neural networks as function approximators reduce the sample complexity in very high dimensional...