“From vehicle dynamics to human-robot interactions ”
Dr. Mauro Da Lio, University of Trento
Friday, October 7, 2016 2:00 – 3:00PM EEB 248
Abstract: This seminar will provide a panorama of the main ideas of my main research line. I will start with the original problem of assessing the maneuverability of unstable vehicles – such as motorcycles – which was solved by imagining that these vehicles were driven “optimally”. I will then introduce the (not surprising) discovery that minimum time optimal control of motorcycles matches the way trained race drivers actually drive. I will then shift to the problem of modeling which optimality criterion holds for ordinary drivers, introducing some theories about optimality of human control (in particular minimum jerk). I will show the use of Optimal Control to model ideal ordinary drivers and its application to produce “reference maneuvers” used as gold standard in Advanced Driver Assistance Systems, showing, in particular, the application of these ideas in the PReVENT project. I will then touch the problem of modelling human driver behavior when multiple choices are possible, reviewing the Simulation Hypothesis of Cognition and its implication for the inference of intentions of drivers (a process that in natural cognition is called mirroring – from mirror neuron theory – and which can also be considered as a “mother nature” version of model-based state estimation). I will show, in particular, the application of this mechanism in the EU InteractIVe project. I will review the notions of subsumption and layered control architecture and, in particular, the role of action selection, and finally introduce the notion of artificial (co)driver agents showing the current status of such an agent for the EU AdaptIVe project. In the conclusions I will introduce the Dreamn4Cars project and the related idea of using dream-like mechanisms to train the sensory-motor architecture of an artificial driver for rare situations developed as variations of real world (near miss) events.
Bio: I am professor of mechanical systems with the University of Trento, Italy. My initial research activity was focused on modeling and simulation of mechanical multibody systems, and on methods for generating the equations of motions symbolically. In particular, I developed symbolic models for vehicle and spacecraft dynamics and – exploiting the availability of symbolic equations – I worked on model-based control, and in particular Optimal Control. This was initially used to study the maneuverability and handling of motorcycles (which are unstable vehicles that cannot be studied otherwise in open loop) and later extended to the modeling of “optimal” drivers. More recently my focus shifted to the modeling of human sensory-motor control, in particular drivers and motor impaired people. In this framework, optimal control motor primitives are part of layered control architectures that can reproduce (to some extent) complex cognition and action-selection processes of humans. According to recent theories this in turn enables several possibilities for human-robot interactions. Prior of academic carrier I worked for an off-shore oil research company in underwater robotics (an EU EUREKA project). I have been involved in several EU framework programme 6 and 7 projects (PReVENT, SAFERIDER, interactIVe, VERITAS, adaptIVe, No-Tremor) in the domains of Intelligent Vehicles and Virtual Physiological Humans. I am currently the coordinator of the EU Horizon 2020 Dreams4Cars Research and Innovation Action: a 4.3 M€ collaborative project in the Robotics domain which aims at increasing the cognition abilities of artificial driving agents by means of an offline simulation mechanism broadly inspired to the dream states.