http://www.nelsonrobotics.org/paper_archive_nelson/nelson-jras-2009.pdf

Fitness functions in evolutionary robotics: A survey and analysis

The reality gap, which often makes controllers evolved in simulation inefficient once transferred onto the physical robot, remains a critical issue in evolutionary robotics (ER). We hypothesize that this gap highlights a conflict between the efficiency of the solutions in simulation and their transferability from simulation to reality: the most efficient solutions in simulation often exploit badly modeled phenomena to achieve high fitness values with unrealistic behaviors. This hypothesis leads to the transferability approach, a multiobjective formulation of ER in which two main objectives are optimized via a Pareto-based multiobjective evolutionary algorithm: 1) the fitness; and 2) the transferability, estimated by a simulation-to-reality (STR) disparity measure. To evaluate this second objective, a surrogate model of the exact STR disparity is built during the optimization. This transferability approach has been compared to two reality-based optimization methods, a noise-based approach inspired from Jakobi's minimal simulation methodology and a local search approach. It has been validated on two robotic applications: 1) a navigation task with an e-puck robot; and 2) a walking task with a 8-DOF quadrupedal robot. For both experimental setups, our approach successfully finds efficient and well-transferable controllers only with about ten experiments on the physical robot.

/pdf/the-transferability-approach-crossing-the-reality-gap-in-3b4rzkied2.pdf

The Transferability Approach: Crossing the Reality Gap in Evolutionary Robotics

Designing agile locomotion for quadruped robots often requires extensive expertise and tedious manual tuning. In this paper, we present a system to automate this process by leveraging deep reinforcement learning techniques. Our system can learn quadruped locomotion from scratch using simple reward signals. In addition, users can provide an open loop reference to guide the learning process when more control over the learned gait is needed. The control policies are learned in a physics simulator and then deployed on real robots. In robotics, policies trained in simulation often do not transfer to the real world. We narrow this reality gap by improving the physics simulator and learning robust policies. We improve the simulation using system identification, developing an accurate actuator model and simulating latency. We learn robust controllers by randomizing the physical environments, adding perturbations and designing a compact observation space. We evaluate our system on two agile locomotion gaits: trotting and galloping. After learning in simulation, a quadruped robot can successfully perform both gaits in the real world.

Sim-to-Real: Learning Agile Locomotion For Quadruped Robots

There is growing need for software tools that can accurately simulate the complex dynamics of modern robots. While a number of candidates exist, the field is fragmented. It is difficult to select the best tool for a given project, or to predict how much effort will be needed and what the ultimate simulation performance will be. Here we introduce new quantitative measures of simulation performance, focusing on the numerical challenges that are typical for robotics as opposed to multi-body dynamics and gaming. We then present extensive simulation results, obtained within a new software framework for instantiating the same model in multiple engines and running side-by-side comparisons. Overall we find that each engine performs best on the type of system it was designed and optimized for: MuJoCo wins the robotics-related tests, while the gaming engines win the gaming-related tests without a clear leader among them. The simulations are illustrated in the accompanying movie.

/pdf/simulation-tools-for-model-based-robotics-comparison-of-utm92xepcx.pdf

Simulation tools for model-based robotics: Comparison of Bullet, Havok, MuJoCo, ODE and PhysX

Interactive rigid body simulation is an important part of many modern computer tools, which no authoring tool nor game engine can do without. Such high-performance computer tools open up new possibilities for changing how designers, engineers, modelers and animators work with their design problems. This paper is a self contained state-of-the-art report on the physics, the models, the numerical methods and the algorithms used in interactive rigid body simulation all of which have evolved and matured over the past 20 years. Furthermore, the paper communicates the mathematical and theoretical details in a pedagogical manner. This paper is not only a stake in the sand on what has been done, it also seeks to give the reader deeper insights to help guide their future research.

Interactive Simulation of Rigid Body Dynamics in Computer Graphics

We present a qualitative evaluation of a number of free publicly available physics engines for simulation systems and game development. A brief overview of the aspects of a physics engine is presented accompanied by a comparison of the capabilities of each physics engine. Aspects that are investigated the accuracy and computational efficiency of the integrator properties, material properties, stacks, links, and collision detection system.

/pdf/evaluation-of-real-time-physics-simulation-systems-4c39p63w4z.pdf

Evaluation of real-time physics simulation systems

Evaluation of real-time physics simulations systems

This article describes WAMbot, a multi-robot system designed to navigate, explore and map large-scale urban environments while performing visual object recognition. The system can be easily adapted and deployed into a variety of scenarios such as search and rescue or surveillance. The original WAMbot system was developed for the Multi Autonomous Ground-robotic International Challenge in 2010 (MAGIC 2010). It successfully passed a competitive down-selection process and placed 4th in the final challenge. Our ongoing work has been to modularize the WAMbot system further, and standardize it using the open source Robot Operating System (ROS) software framework. We describe the core components, or stacks, in the new WAMbot system including: 1) a large-scale mapping stack capable of building 500×500m global maps from over 20 robots in real-time, 2) an efficient, frontier-based multi-robot exploration stack, 3) a navigation stack including a centralized global planner with decentralized local navigation and 4) an intuitive graph-based visual object recognition pipeline.

Cooperative multi-robot navigation, exploration, mapping and object detection with ROS

We propose a novel approach for transferring evolved control systems from a simulated environment to a real robot. Multiple dynamic simulation systems are simultaneously employed to provide a valid range of simulation variance that can be exploited to generate robust controllers in a purely virtual environment. These controllers can then be directly transferred to a physical robot.

Adrian Boeing

Papers

Evaluation of real-time physics simulation systems

Evaluation of real-time physics simulations systems

Cooperative multi-robot navigation, exploration, mapping and object detection with ROS

Leveraging multiple simulators for crossing the reality gap

An implementation of a garment-fitting simulation system using laser scanned 3D body data