Title: Autonomous drone racing and footstep planning for quadruped robots
Abstract: Drones, especially quadrotors, have shown their great value for applications like aerial photography, object delivery and warehouse inspection. At the same time, with the development of Artificial Intelligence (AI), computers can replace humans and even perform better than humans in some areas where it was impossible before like the AI program Alpha Go which beat the human world champion in Go matches and Alpha star which was rated above 99.8% human players in the real-time strategy game StarCraft II. Concerning drones, the question is whether they can fly races completely by themselves and if they can fly even faster than human pilots’ racing drones?
When humans walk in uncertain environments, they won't step on ‘unsafe’ terrain directly. Instead, humans will either use their feet to touch the terrain to confirm the safety or make a detour. We want to teach our quadruped robots this strategy to make them walk in uncertain environments in a safer way.
Kevin Sebastian Luck
Title: Data-Efficient Co-Adaptation of Robots with Graph Neural Networks
While much progress was made in recent years in both the evolutionary robotics and the deep learning community regarding the adaptation of robot morphology, no strong emphasis was and is placed on the data-efficiency of the developed approaches. Newly developed methodologies are primarily evaluated in simulation and make use of the current abundance of computational resources and mass-parallelization of simulation. This limits the applicability of these technologies to the co-adaptation of robots in the real world.
In this talk I will present our current work and preliminary results on a framework capable of adapting robot morphologies and their behaviour in a data-efficient way which is suitable for a future deployment in the real world. This approach makes use of a close inter-connection between deep reinforcement learning on graph neural networks and evolutionary optimization to co-adapt robots in a fast and sample-efficient manner.