Although robotic systems must ultimately be demonstrated in practice, they can rarely be developed, tested and evaluated using real data. Collecting data for real systems is both complex and time-consuming, as there may be limited access to platforms or sensors. Furthermore, issues of safety and practicality mean that empirical data is often collected for environments which are less massive, complicated or dynamic than the target environments. One means of overcoming these limitations is to evaluate systems in simulation, provided of course that it a reasonable representation of reality.
Significant recent progress has been made in experimental robotics software, and a number of community-backed open source projects exist. However, contributions tend to be added in a piecemeal fashion, and in response to specific research project needs. Consequently, little emphasis has been placed on modelling general environmental conditions such as wind, non-Gaussian noise models for sensors, and general dynamics / control models for standard vehicles. As a result of a lack of sensor and control interface standardisation, in the general case it remains challenging to transition reliably between simulation and hardware, as well as carry out experiments on heterogeneous robots.
This workshop is organised around three themes that make up robotic simulation — Environmental and Sensor Noise Models, Vehicle Dynamics and Control Models, and Architectures, Systems and Experiences. The overarching objective of the workshop is thus to:
- Disseminate information about state of the art models and software tools that facilitate realistic, rapid and repeatable robotic experiments.
- Discuss aspects of robot simulation that require improvement, with the objective of formulating a plan to develop what is needed.
- Provide a forum in which developers can meet with researchers to discuss how models can be implemented in a way that promotes reuse.
One of the distinguishing characteristics of this workshop is that it provides a forum in which developers — from both industry and from non-profit organisations — can meet with academics to discuss and shape the future of robot simulation. In order to promote a pragmatic view of simulation, this workshop features two data-centric, hands-on sessions with state-of-the-art simulation tools. The ultimate goal is to provide a suitable theoretical and practical platform from which a constructive and holistic discussion of robotic simulation can emerge.
The workshop is purposefully aimed at a wide audience. Firstly, for those who currently only conduct hardware experiments as part of their research and development cycle, this workshop will show that — with a suitable model — simulation may be quicker to carry out, allow one to evaluate over a wider range of conditions, and be lighter on overall resources. Secondly, this workshop will prompt those already using simulation to think about whether the environment, sensor, dynamics and control models they use are strongly reflective of reality — are they perturbed by noise, and is the noise distribution realistic? Finally, this workshop will provide a forum for those already using complex simulations noise models to discuss how to refine and extend those models.