Community Research and Development Information Service - CORDIS


CON-HUMO Report Summary

Project ID: 337654
Funded under: FP7-IDEAS-ERC
Country: Germany

Mid-Term Report Summary - CON-HUMO (Control based on Human Models)

CON-HUMO focuses on novel concepts of automatic control based on data-driven human models and machine learning. The envisioned scientific and technological developments enable innovative control applications that are difficult to achieve with traditional control methods particularly in the area of human-robot interaction (HRI) in which the derivation of the model with classical system identification techniques is difficult because of the complexity and probabilistic nature of human behavior.
We have made significant advancements towards i) data-driven Bayesian modeling of human motion behavior, ii) systematic control design for this type of models, and iii) its implementation to achieve intuitive and safe HRI. Our focus has been on stochastic modeling approaches to capture inherent human variability. In particular, we focused on Gaussian Process (GP) models for their benefits to not only provide a prediction but also a measure of model confidence depending on the distance to the training data. We developed a novel mathematical framework for GP regression for rigid body motions, i.e. over the Special Euclidean group SE(3). Based on our results, it is now possible to predict the human motion (or of any other rigid body) including a prediction uncertainty certificate from a GP model. In addition, we achieve more accurate predictions through the incorporation of neuroscientific findings into the model structure. Furthermore, we have developed systematic control approaches based on data-driven Bayesian models. We investigated a class of stochastic optimal control approaches for robotics that explicitly incorporates the prediction uncertainty of GPs and other probabilistic representations into the control design. We have demonstrated that this type of controls significantly improves the human-perceived quality of robotic assistance. Furthermore, we developed a constrained control approach, which is crucial for guaranteeing safety in HRI. All control designs are accompanied by a rigorous analysis, i.e. we can prove their properties in terms of stability and constraint satisfaction. Our achievements in a nutshell: Our results enable a more accurate prediction of human motion behavior, which is crucial for intuitive HRI. We have also gained a deeper understanding of human-human interaction, which is an important precursor to generate intuitive robot behaviors. Furthermore, we performed the first important step towards a systematic model-based control design for intuitive and safe HRI based on human models. Notably, our fundamental results on systematic control design based on data-driven non-parametric models with formal guarantees translate far beyond the application in HRI.
CON-HUMO is strongly interdisciplinary with its ambition to bring neuroscientifically inspired data-driven models of human behavior into control engineering. The realization of close human-machine interaction has been approached from both perspectives. The cross-disciplinary research developments have been capitalized by recruiting personnel with complimentary backgrounds including control engineering, mathematics, and neuroscience. We have performed knowledge and technology transfer through publication of our results in conferences/workshops and journal papers.
Bringing advanced smart-technology into the society can be a sensitive and ethically challenging issue to the general public. We therefore regularly display our research activities and outcomes not only in the relevant academic groups, but also to the local students and interest groups in the public. For instance, we host a research day to open our laboratory for current / prospective students of the host University at least twice a year. Furthermore, workshops have been organized to external non-academic groups / organizations to appeal general interests in control concepts and robotic technologies. In the first reporting period, we presented our works to over 70 visitors from the academia, industries, local groups / societies at our laboratory. As an application of our research developments, we have been in contact with a local medical hospital to study how the modeling techniques developed in CON-HUMO can be adapted to technology-led support systems in clinical settings as part of rehabilitation and/or a diagnostic tool.


Ulrike Ronchetti, (Legal Representative)
Tel.: +498928922616
Fax: +498928922620
Record Number: 193266 / Last updated on: 2017-01-17
Follow us on: RSS Facebook Twitter YouTube Managed by the EU Publications Office Top