Invited Speakers

 

john

John Leonard

John J. Leonard is Professor of Mechanical and Ocean Engineering in the MIT Department of Mechanical Engineering and a member of the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL). His research addresses the problems of navigation and mapping for autonomous mobile robots. He holds the degrees of B.S.E.E. in Electrical Engineering and Science from the University of Pennsylvania (1987) and D.Phil. in Engineering Science from the University of Oxford (formally 1994). Prof. Leonard joined the MIT faculty in 1996, after five years as a Post-Doctoral Fellow and Research Scientist in the MIT Sea Grant Autonomous Underwater Vehicle (AUV) Laboratory.

Simulations are Doomed to Succeed: Thoughts on Simulation for Mobile Robot Navigation and Mapping

Abstract — In this talk, I will reflect on the dangers of relying too much on simulations in robotics, but I will also discuss some successes that have been enabled by simulations for SLAM and for Marine Robotics. I will also present some simulation challenges for the future in relation to developing robust Self-Driving Vehicles.

emo

Emo Todorov

Emanuel Todorov is an Associate Professor in the College of Engineering at the University of Washington. He graduated from MIT in 1998 with a PhD in Cognitive Neuroscience. He joined the University of Washington from the Department of Cognitive Science at the University of California San Diego. His research interest lies in the control of complex movements in animals and robots.

Comparison of physics engines for robotics

Abstract — 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 will describe quantitative measures of simulation performance, focusing on the numerical challenges that are typical for robotics as opposed to multi-body dynamics and gaming. We will then present simulation results obtained by instantiating the same model in multiple engines and running side-by-side comparisons. The overall conclusion is that each engine performs best on the type of system it was designed and optimized for. Our engine (MuJoCo) wins robotics-related tests, while gaming engines win multi-body dynamics tests, with ODE being somewhat better than Bullet, Havoc and PhysX. We will also describe the new model of contact dynamics used in MuJoCo.

michael

Michael Gschwandtner

Michael Gschwandtner is currently working as a Junior Scientist at the Austrian Institute of Technology (AIT) in the field of automated document authentication. In 2013 he completed a PhD at the University of Salzburg as a member of the Wavelab group, working in the field of computer vision and image processing. One of Michael’s key contributions is Blensor — an Open Source Simulation Package for Light Detection and Ranging (LIDAR/LADAR) and Kinect sensors.

Software based sensor simulation — Lowering the barrier for robotics research

Abstract — Over the last few years the number of sensors for acquiring three dimensional data has exploded, ranging from devices for a few hundred dollars to several tens of thousands. This has made the field of robotics more accessible than ever but naturally the different types of sensors have their individual benefits and limitations. This is one of the key aspects where simulating different types of sensors is a helpful decision-making tool for acquiring the right type of sensor. But the use cases for sensor simulation go far beyond a mere decision-making tool. Simulating impossible or unlikely scenarios, cost reduction via sensor setup evaluation before running long term scans, unit testing and of course education are all examples for scenarios where simulation plays a key role. This talk will give an overview over the different use cases and explain when to use software simulation and when the real sensor should be preferred.

girish

Girish Chowdhary

Girish Chowdhary is an assistant professor at Oklahoma State University, and the director of the Distributed Autonomous Systems laboratory at OSU. He holds a PhD from Georgia Institute of Technology, and has postdoctoral experience at the Laboratory for Information and Decision Systems (LIDS) of the Massachusetts Institute of Technology for about two years.  Prior to joining Georgia Tech, he also worked at the German Aerospace Center’s (DLR’s) Institute of Flight Systems for around three years.  Girish’s ongoing research interest is in theoretical insights and practical algorithms for adaptive autonomous decision making over massive spatiotemporal scales, with a particular focus on applications in Unmanned Aerial Systems. He is a recipient of the AFOSR Young Investigator Award.

Hardware-in-the-Loop simulations for UAS Research

Abstract:  Flight-testing is critical in providing confidence in the reliability and effectiveness of unmanned aerial systems (UAS) autonomy algorithms in real-world environments. However, restrictive regulations from federal agencies, and the cost of flight-testing in general, can make it difficult to engage in an extensive flight testing campaign. In this talk, I will present an overview of Hardware-in-the-Loop (HITL) simulation architectures. HITL simulations are designed to tightly couple flight computational and sensing hardware with high-fidelity physics based UAS and environment models to create a graphic and highly realistic virtual flight-testing environment. Through HITL simulations, researchers can test the software and hardware systems in a low cost and zero-risk environment before engaging in costly flight-testing. I will demonstrate how off-the shelf gaming platforms, such as X-plane, can be utilized to quickly create usable HITL simulation environments at the scale of an academic laboratory. I will end the talk with new horizons in multi-agent HITL simulations. In particular, I will talk about our newly developed Multi-Agent Game Emulator (MAGE) HITL environment designed to enable multiple aircraft to interact together in a common cloud based environment. MAGE is inspired by massive online games, such as the World of Warcraft, and provide a highly capable and unique perspective in engaging in multi-agent aerospace autonomy research.