Author: Jean-Philippe Diguet, CNRS Research Director at Lab-STICC, Lorient / Brest, France, visiting Prof. at USP/ICMC.

Abstract: The talk will address the interest of using self-adaptive system-on-chip in the context of autonomous system, it will present current and future challenges including the integration in a SOA approach.
Future autonomous vehicles (UAV, AUV, etc.) will be designed for conducting missions with human interactions reduced to what is really necessary. A mission means reaching objectives according to random events and environment variations. On the one hand this autonomy requires embedded high performance computing (HPC) in order to (i) sense the world by means of sensors such as cameras, radars etc. and to (ii) make decisions with a strategy including a balanced risk and QoS management. But on the other hand HPC means important power consumption and payload (storage, fans, etc.) and this is even more important in SWARMS where the aim is to design multiple low-cost autonomous systems. A promising approach, to address this issue, is to take advantage of reconfigurable hybrid chip (FPGA / Processor) in order to maximize energy-efficiency (MIPS per Watt) at runtime. The computing and energy efficiency can be obtained by means of dedicated and parallel hardware accelerators that are configured according to Mission requirements. Because of scarce embedded resources, the hardware and software resources must be shared according to priority and efficiency optimization. Thus, the autonomous system-on-chip must decide hardware and software implementations depending on the mission status, namely the objective (e.g. search and rescue), the time constraints (e.g. camera data-rate / UAV speed) and the environment (e.g. wind, flying / static obstacles, empty/inhabited areas). According to the mission status that evolves from time to time, the allocation of hardware will change while being shared by mission (e.g. pattern recognition), navigation (e.g. path planning) and safety (e.g. collision avoidance) tasks. Last but not least, autonomous vehicles require a high level of reliability; it means that the management of hardware and software reconfigurations must be as safe as embedded software for critical systems.
The HPeC project addresses this question by considering three levels: 1) the Mission management based on stochastic techniques, 2) the Hardware/Software configuration based on formal Discrete Controller Synthesis and 3) The underlying flexible architecture that relies on a tile-based reconfigurable processor.
The related main challenges are 1) Embedded Intelligence vs Limited Resources, 2) Reliability vs Dynamic reconfiguration and 3) Performance vs Programmability vs Flexibility.
Regarding the programmability challenge and the evolution towards SWARMs of heterogeneous autonomous-systems sharing resources over the air, SOA approaches appear as a very interesting approach. Such approaches are developed at USP/ICMC, this is an opportunity for an international collaboration that also raises a new challenge, which is the use reconfigurable hardware without loss in terms of performance and flexibility.

Short Bio:
Jean-Philippe Diguet is a CNRS director of research at Lab-STICC, Lorient/Brest, Brittany, France. He received the Ph.D degree from Rennes University (France) in 1996. He has then been a visitor researcher at IMEC (Belgium), an associate professor at UBS University (France) and he co-funded in 2003 the dixip company in the domain of wireless embedded systems. Since 2004 he is a full CNRS researcher at Lab-STICC, where he is heading the MOCS team. He has been a visitor researcher at the University of Queensland, Australia in 2010 and an invited Prof. at Tohoku University, Japan in Nov. 2014. His current work focuses on various aspects of embedded system design: i) Self-adaptive System on Chip, ii) Smart Network-On-Chip, iii) CAD tools for SoC synthesis, iv) Dedicated hardware architectures including security and embedded vision. In 2016 he is heading the French project “HPeC” targeting efficient embedded systems for UAV and initiates the collaborative project “SAMM” about autonomous vehicles in marine environment.