Distributing Intelligence in IoT

An increasing number of actuators is being integrated into the IoT, ranging from mobile and wearable devices, over home automation systems to assistance, companion and industrial robots. These actuators must be provided with the necessary intelligence to tackle the complexity, dynamism and uncertainty in their operational contexts.

Our current key research interests are:

Deep learning on resource-constrained devices

Cloud-based machine learning is computationally demanding. We research how to implement these techniques to process the information on-board, for example on the embedded GPU or its neuromorphic chip. This research involves a trade-off between computing power and fidelity, which should not necessarily be done at design time.

Platform support for distributed intelligence and edge cloud

Developers cannot anticipate to all possible solutions at design time. We research and develop the necessary platform solutions to provide interoperability for real-time discovery, negotiation and on-the-fly allocation of additional resources and services: on-board, on nearby devices or on edge cloud infrastructure.

IoT-assisted and cloud robotics

Open worlds and human presence requires advanced capabilities to observe, understand and interpret the surroundings and dynamically adjust a robot’s control strategy accordingly. We research how robot task planning and execution can be improved by leveraging on sensing and actuation functionality provided by other agents in the nearby environment, and by relying on the cloud for computational offload or as knowledge database.

Intelligent compliant mechatronics

We explore the possibility to (partially) outsource functionality and computational tasks to compliant body elements of the robot which drastically reduces the complexity of the motor control. The main objective of this research theme is to derive a set of precise design rules for constructing a well performing morphology for the task of robot control and cognition.

IDLab examples

In the WONDER project, we investigated the requirements for a companion robot to work 24/7 and semi-autonomously in nursing homes. This implied careful planning of the robot tasks over the day, taking into account battery, walking speed and context information.

This implies that the robot needs to be able to move from one resident to another and start interacting personally, based on a resident’s individual profile. Taking into account hardware limitations (battery, walking speed), we have researched how to ideally schedule intervention strategies throughout the day.

In the CLAXON project, we have developed and deployed a multi-sensor robot that works seamlessly and safely with humans on a car assembly line. For more details, see our case study with Robovision.

The DYAMAND middleware offers an easy and practical way of creating, deploying and managing smart applications in today’s jungle of devices running in home, office, IoT and industrial networks.

DIANNE is an open-source modular software framework for designing, training and evaluating artificial neural networks. It is built on top of OSGi and Aiolos and can transparently deploy and redeploy (parts of) a neural network on multiple machines, as well as scale up training on a compute cluster.