Keynote Lectures

Soft Evolutionary Robotics: Adapting morphology and control to the world around us
Andre Rosendo, ShanghaiTech University, China, China

ATLASCAR: a sample of the quests and concerns for Autonomous Cars
Vitor Santos, Universidade de Aveiro, Portugal, Portugal

Deep Learning for Robot Navigation and Perception
Wolfram Burgard, University of Freiburg, Germany, Germany

Robotic Dressing Assistants: Research, Ethics and Fiction
Carme Torras, , Spain, Spain

Abstract
Biological forms started primitively and, eventually, some of these forms evolved and reached intelligence. Although current artificial intelligence (AI) explores learning techniques to create thinking machines, is “thinking” really a condition for “existence”? Instead of creating intelligent computer programs to later interact with the physical world, my research will start with physical interactions and create intelligence from there, as achieved by humans. I will present our latest results with Mother Robot, a robot capable of creating children robots, which are smaller robots with the capacity of growing their bodies according to cues from the interaction with the environment. Additionally, I will talk about robots capable of simultaneously altering their morphology and control and compare this method with other design ramifications, such as solely altering control parameters. The future of robotics remains uncertain, but the capacity of adapting their self-design and altering their morphology to fulfill tasks more effectively can be a driving force for future Evolutionary Robotics.

Abstract
Autonomous cars have become an irreversible trend in the modern world society from several points of view, and will be one of the most relevant topics in mobility and transportation in the near and long term future. However, despite the many solutions and prototypes that have been rising in a growing set of countries in the last years, no definitive and universally usable solution has been made available by automotive companies or research laboratories. The issues of vehicle control are perhaps the best solved due to a long term research in engineering and actuators technology, but what still remains as a challenge is the reliable awareness of the vehicle in what concerns the dynamic and complex environment where it is embedded, namely, in urban scenarios. This concern is related mainly to robust and reliable perception and data processing.

ATLASCAR is a project born in 2010 where many engineering issues had to be solved, but that represent only by themselves too little as far as pushing forward the state-of-art is concerned. What possibly emerged from this project as contributions useful to other authors and researchers are the options in the system architecture, but mainly the study of a set of problems less frequently addressed, but that can make a difference in the long term intelligent vehicles. Some simpler examples are proprioceptive issues such as the automatic calibration of multi-modal sensors or the measurement with high precision of the vehicle orientation on the road plane, up to concerns of pedestrian posture detection to predict their behaviour, or how to automatically merge into the traffic flow with other moving vehicles. Pushing the challenge even further, assess the driving typology or characteristics of the driver navigation behaviour may point to a new level of driving assistance. Actually, although autonomous driving is sought and desired by many, it is expectable that future cars will allow at least two control modes: automatic or manual with advanced assistance systems. So, this ability to assess and monitor the driver behaviour and driving typology may open the way to some sort of predictive and adaptive driver assistance systems, which is so, or even more important than the autonomous driving capabilities of the car.

Abstract
Autonomous robots are faced with a series of learning problems to optimize their behavior. In this presentation I will describe recent approaches developed in my group based on deep learning architectures for different perception problems including object recognition and segmentation and using RGB(-D) images. In addition, I will present a terrain classification approaches that utilize sound and vision. For all approaches I will describe extensive experiments quantifying in which way the corresponding approaches extend the state of the art.

Abstract
The versatile manipulation of clothing items by robots would considerably enlarge their range of application, from textile manufacturing to online shopping, housekeeping, hospital logistics, and providing dressing assistance to elderly and disabled people. To carry out these tasks in unprepared human environments, robots need to be easily instructed by non-experts, intrinsically safe to people, capable of communicating and working collaboratively, and highly adaptable to changing situations and new goals. A quick overview of the research challenges arising in this context will be illustrated through projects being developed at the Perception and Manipulation Lab of IRI (CSIC-UPC).

Robotic assistants pose also fundamental ethical questions, many practical ones stemming from autonomous robot decision making conflicting with human freedom and dignity. Philosophy, psychology and law are shedding principled light on these conflicts, while science fiction permits freely speculating about possible future scenarios, thus encouraging debate on the pros and cons to try to guide techno-scientific research in the most desirable direction.