Today we live in a context in which devices are increasingly interconnectedand sensorized and are almost ubiquitous. Deep learning has become in recentyears a popular way to extract knowledge from the huge amount of data thatthese devices are able to collect. Nevertheless, centralized state-of-the-artlearning methods have a number of drawbacks when facing real distributedproblems, in which the available information is usually private, partial,biased and evolving over time. Federated learning is a popular framework thatallows multiple distributed devices to train models remotely, collaboratively,and preserving data privacy. However, the current proposals in federatedlearning focus on deep architectures that in many cases are not feasible toimplement in non-dedicated devices such as smartphones. Also, little researchhas been done regarding the scenario where data distribution changes over timein unforeseen ways, causing what is known as concept drift. Therefore, in thiswork we want to present Light Federated and Continual Consensus (LFedCon2), anew federated and continual architecture that uses light, traditional learners.Our method allows powerless devices (such as smartphones or robots) to learn inreal time, locally, continuously, autonomously and from users, but alsoimproving models globally, in the cloud, combining what is learned locally, inthe devices. In order to test our proposal, we have applied it in aheterogeneous community of smartphone users to solve the problem of walkingrecognition. The results show the advantages that LFedCon2 provides withrespect to other state-of-the-art methods.