Gazebo Retrospective


The DARPA Robotics Challenge (DRC) has come to a successful close with impressive performances by all the competitors. Youtube has all the highlights and runs, in case you missed the excitement.

At the outset of the DRC, Program Manager Dr. Gill Pratt said, “The value of a cloud-based simulator is that it gives talent from any location a common space to train, design, test and collaborate on ideas without the need for expensive hardware and prototyping. That opens the door to innovation.” Taking that goal to heart, we worked together with DARPA and the DRC teams to make Gazebo into the most flexible, powerful, and capable robotics simulation available.

A few key milestones that have been achieved over the past three years include:

Real-time high-fidelity simulation of a humanoid robot

The Atlas robot from BDI, a 53-degree-of-freedom sensor heavy humanoid robot, runs on a modern desktop computer at or near real-time. To our knowledge, this represents the most complex robot simulation to date that runs in near real time on commodity computing hardware.

Support for four different physics engines

In addition to ODE, Gazebo supports DART, Simbody, and Bullet. This is the first time that such a diverse set of physics engines has been supported in one simulator.

Integration of aerodynamics and hydrodynamics

Gazebo’s flexible plugin system has made it possible to simultaneously simulate ground, aerial, and underwater vehicles in expansive environments. Physically accurate simulation in all three environments is a first for a robotics simulator without suppressing physics properties.

Cloud-hosted global simulation competition

As a preliminary stage in the DRC program, a virtual competition was conducted to determine which teams would receive funding and access to an Atlas humanoid robot. Leveraging advances in robot simulation, cloud hosted infrastructure, and web technology, we were able to accurately and efficiently simulate complex robots and environments on remote servers while providing realistic data streams for human-in-the-loop robot control.

The competition was a resounding success, with no major problems, and largely positive feedback from the participants. Importantly, the VRC achieved its goals: the most capable software teams were selected for award of an Atlas robot, and those teams were prepared immediately upon receipt of their Atlas, because they had already spent months developing their software using a high-quality simulation of the robot.

Use of Gazebo has expanded beyond the DRC and even the research community. Organizations such as FIRST and RoboCup have begun to use Gazebo either officially or experimentally. Industry is also adapting Gazebo. The Amazon Picking Challenge, a competition designed around complex object manipulation using a robot platform, utilizes Gazebo as a development and testing tool. Rethink Robotics simulates their Baxter industrial robot in Gazebo, and the ROS-Industrial Consortium actively makes use of Gazebo to simulate industrial arms.

Dr. Pratt said, “The DRC Simulator is going to be one of DARPA’s legacies to the robotics community.” We believe that goal has been achieved, thanks in large part to the support and contributions of an active community of simulation developers and users. We are excited to help provide users with this critical tool, and look forward to advances that simulation will take in the coming years.