Intelligent Autonomous Robots and Systems (IAR&S)

Bringing revolutionary benefits to society, while facing important societal challenges
orlando.avila's picture

Orlando Avila GarcíaSoftware Architect

Description / Emerging Technologies: 

Intelligent Autonomous Robots and Systems Research Line coordinates ARI research and innovation activity related to autonomous systems and intelligent automation, be it research and innovation on the technologies themselves or their application in specific vertical markets, or in relation to other technologies or solutions.

We are interested on the kind of systems designed to automate (replace) human decision-making and problem-solving capabilities. Intelligent autonomous systems encompass cyber-physical systems which attain goals in open-ended physical environments:

  • they are self-sufficient (able to gather needed resources) and situated (engaging the environment through their own sensors and actuators);
  • they display a high degree of autonomy to make decisions and take actions;
  • they continuously adapt to a complex, dynamic and uncertain environment;
  • they (collaborate to) solve human physical or mental tasks (goal-directed behaviour); and
  • they self-learn (build knowledge) from their experience in a continuous interaction with the environment.

The challenges we address, and current technological trends are the following:

Adaptive Systems – We are interested in a biologically-inspired and bottom-up approach to AI (Artificial Intelligence) and the kind of intelligent behaviour emerging from the interaction of an embodied agent within a dynamic uncertain complex environment.

Autonomous Systems – We work on decisional autonomy, situational awareness and action selection in cyber-physical systems spanning from wheeled robots to drones, taking a behaviour-based robotics (reactive planning) approach.

Cognitive Robotics – Cognitive neuroscience has progressively provided insights about the kind of physiological mechanisms underlying affective and cognitive phenomena in humans and animals. We take inspiration from it to depart from more engineered and symbolic (top-down) approaches to AI and address with guarantees challenges such as human-robot interaction and social robotics.

Development and Epigenetic Robotics – We are interested in the kind of agents which acquire (implicit) knowledge from its experience in a continuous interaction with the environment, that is, beyond pre-defined knowledge bases and rules engineered by humans. We exploit and combine concepts and methods such as self-organization, collective intelligence, deep learning and reinforcement learning.

Robotic Systems Development – Successful real-world autonomous cyber-physical systems face important challenges in their design, construction, verification and validation, deployment and evolution. We work on the combination of traditional software and systems engineering concepts, methods and tools with new AI systems development ones. We put special focus on MBSE (Model-Based Software Engineering).

Robot and Robotic System Safety - AI safety is a rapidly evolving field that has become one of the key pillar for a successful deployment of intelligent autonomous systems. We are interested in challenges such as robot ethics, safety by design and human-robot interaction safety; specifically, on affective mechanisms to embed safety guarantees in intelligent autonomous systems for operation in proximity to and/or in collaboration with humans.

Critical Mass / Market: 

AI-based Cyber-Physical Systems (CPS) are intelligent robotics systems, or technical systems of networked computers, robots and artificial intelligence, that interact with the physical world to attain a goal. They can be characterized as intelligent robotics systems interacting with the physical world, connected to their environment through the IoT, and getting unlimited computational, data and networking resources from the cloud. Assuring the autonomy of those systems (correctness, robustness, reliability, safety, etc.), which are designed to continuously adapt (e.g. by learning) to its environment, possesses important challenges to developers and roboticists.

  • E-FLY: Edge-enabled Flying Inspector (EIT Digital, 2019-2020). E-FLY builds a distributed UAV (unmanned aerial vehicle), edge and cloud AI image processing framework for analysis of infrastructures. Responsibility: collision warning system and collision avoidance assistance to UAV operators.

Cloud Robotics and Decentralized AI. AI-based CPS designers and vendors are under pressure to provide increasingly intelligent behaviour at competitive costs. Increasing on-board computation of robots raises their costs and energy demand while reducing autonomy. Cloud Robotics’ aim is to bring cloud technologies in terms of processing power, storage and services to extend robots capabilities. Even if powerful, drawbacks recognized for Cloud robotics approach relate to the lack of adequacy for tasks requiring real-time processing, low latency and the requirement for continuous internet connectivity. AI decentralization strategies, such as edge computing, have emerged to address those drawbacks.

  • DECENTER: Decentralized technologies for cloud-to-edge intelligence (ICT 2018-2021). The goal of the project is to deliver a robust Fog Computing Platform, covering the whole cloud-to-things continuum, that will provide orchestration and provisioning of federated resources for AI (Artificial Intelligence) applications and systems. Responsibility: connected robots and deployment of robotic tasks at the edge (edge robotics).

Robot Safety. Designing CPSs for operation in proximity to and/or in collaboration with humans means that current safety engineering and legal mechanisms need to be revisited to ensure that individuals –and their properties– are not harmed. AI-based CPSs need to go beyond being fail-safe to become safe-operational, that is, to maintain their behaviour within safety boundaries even during the time it takes to reach a safe state after a fail. Atos is working with the embodied cognition and affective computing communities in AI to embed trust and safety mechanisms in autonomous robots in order to ensure robots remain safe-operational in dynamic, complex and unpredictable environments. 

  • We co-organized and co-chaired the First International Workshop on AI Safety Engineering (WAISE2018) whose purpose is to bring together traditional safety engineering and AI safety communities. We plan to continue celebrating this event in the context of SAFECOMP 2019.
Relation between the Research Line and Atos Portfolio : 
  • Artificial Intelligence (AI) technologies enable a digital future where products and services deliver people-first experiences to customers. AI adds an adaptive layer on top of existing IT and addresses the complexity arising from the interplay between Big Data, IoT and Cyber-Physical Systems (such as autonomous cars).
  • IAR&S Research Line contributes to shaping the future Atos Industry 4.0 and IoT offering. It envisions intelligent autonomous systems and robots that streamline real-world operations in a wide range of industries by means of real-time automated data-driven decisions and actions.