Assessing Transferability From Simulation to Reality for Reinforcement Learning

Instance Weighted Incremental Evolution Strategies for Reinforcement Learning in Dynamic Environments

Evolution strategies (ESs), as a family of black-box optimization algorithms, recently emerge as a scalable alternative to reinforcement learning (RL) approaches such as Q-learning or policy gradient and are much faster when many central processing units (CPUs) are available due to better parallelization. In this article, we propose a systematic incremental learning method for ES in dynamic environments. The goal is to adjust previously learned policy to a new one incrementally whenever the environment changes. We incorporate an instance weighting mechanism with ES to facilitate its learning adaptation while retaining scalability of ES. During parameter updating, higher weights are assigned to instances that contain more new knowledge, thus encouraging the search distribution to move toward new promising areas of parameter space. We propose two easy-to-implement metrics to calculate the weights: instance novelty and instance quality. Instance novelty measures an instance's difference from the previous optimum in the original environment, while instance quality corresponds to how well an instance performs in the new environment. The resulting algorithm, instance weighted incremental evolution strategies (IW-IESs), is verified to achieve significantly improved performance on challenging RL tasks ranging from robot navigation to locomotion. This article thus introduces a family of scalable ES algorithms for RL domains that enables rapid learning adaptation to dynamic environments.

A Dirichlet Process Mixture of Robust Task Models for Scalable Lifelong Reinforcement Learning

While reinforcement learning (RL) algorithms are achieving state-of-the-art performance in various challenging tasks, they can easily encounter catastrophic forgetting or interference when faced with lifelong streaming information. In this article, we propose a scalable lifelong RL method that dynamically expands the network capacity to accommodate new knowledge while preventing past memories from being perturbed. We use a Dirichlet process mixture to model the nonstationary task distribution, which captures task relatedness by estimating the likelihood of task-to-cluster assignments and clusters the task models in a latent space. We formulate the prior distribution of the mixture as a Chinese restaurant process (CRP) that instantiates new mixture components as needed. The update and expansion of the mixture are governed by the Bayesian nonparametric framework with an expectation maximization (EM) procedure, which dynamically adapts the model complexity without explicit task boundaries or heuristics. Moreover, we use the domain randomization technique to train robust prior parameters for the initialization of each task model in the mixture; thus, the resulting model can better generalize and adapt to unseen tasks. With extensive experiments conducted on robot navigation and locomotion domains, we show that our method successfully facilitates scalable lifelong RL and outperforms relevant existing methods.

Robot Learning From Randomized Simulations: A Review

The rise of deep learning has caused a paradigm shift in robotics research, favoring methods that require large amounts of data. Unfortunately, it is prohibitively expensive to generate such data sets on a physical platform. Therefore, state-of-the-art approaches learn in simulation where data generation is fast as well as inexpensive and subsequently transfer the knowledge to the real robot (sim-to-real). Despite becoming increasingly realistic, all simulators are by construction based on models, hence inevitably imperfect. This raises the question of how simulators can be modified to facilitate learning robot control policies and overcome the mismatch between simulation and reality, often called the “reality gap.” We provide a comprehensive review of sim-to-real research for robotics, focusing on a technique named “domain randomization” which is a method for learning from randomized simulations.

Adjustable and Adaptive Control for an Unstable Mobile Robot Using Imitation Learning with Trajectory Optimization

In this contribution, we develop a feedback controller in the form of a parametric function for a mobile inverted pendulum. The control both stabilizes the system and drives it to target positions with target orientations. A design of the controller based only on a cost function is difficult for this task, which is why we choose to train the controller using imitation learning on optimized trajectories. In contrast to popular approaches like policy gradient methods, this approach allows us to shape the behavior of the system by including equality constraints. When transferring the parametric controller from simulation to the real mobile inverted pendulum, the control performance is degraded due to the reality gap. A robust control design can reduce the degradation. However, for the framework of imitation learning on optimized trajectories, methods that explicitly consider robustness do not yet exist to the knowledge of the authors. We tackle this research gap by presenting a method to design a robust controller in the form of a recurrent neural network, to improve the transferability of the trained controller to the real system. As a last step, we make the behavior of the parametric controller adjustable to allow for the fine tuning of the behavior of the real system. We design the controller for our system and show in the application that the recurrent neural network has increased performance compared to a static neural network without robustness considerations.