Machine-to-machine (M2M) is one of the main challenges that will drive the future evolution of wireless cellular networks, already engaged with the 5th generation. M2M communications are mandatory to support new digital twinning applications. The classical cellular infrastructure needs a deep evolution to adapt to this new paradigm: instead of files, or stream, the information is composed of small messages on sensing actions from the real world to its digital twin. The project will focus on non-terrestrial network (NTN) infrastructure required to ensure a full terrestrial coverage. An important property of such M2M networks is that working in connected mode is not feasible: maintaining the connection with myriad nodes would introduce a prohibitive cost. Moreover, classical training sequences used to learn the channel are a waste of resources from a communication point of view and should be suppressed to optimize resources. The goal of the WARM-M2M project is to tackle the problem of truly asynchronous transmission in the context of massive multi-user applications with small payload and sporadic transmission with little feedback channel. The consortium will join their background in 3 fields that are rarely studied together. First, the definition of all-inclusive waveform design (QCSP frames and tensor based frames) to avoid preamble. Second, the advanced multi-user decoding algorithm (hybrid Gaussian approximate message passing and Graph Neural Network) to tackle jointly the problem of active-users identification, channel estimation and decoding through a graph representation of the whole problem, and 3) the use of deep reinforcement learning (DRL) techniques to optimize the protocol layer. A major measure of success of WARM-M2M will be the integration of the proposed solution in future NTN standards.