AI in Energy and Mobility Systems: From Technological Promise to Collective Responsibility

We push our technologies to reach tomorrow ever faster. But when it comes time to hit the road, one question remains: in what vehicles, and with what energy sources, do we wish to move forward?

On November 19th, Toulouse hosted a new edition of Future Intelligence. This year, it addresses a central challenge: the role of AI in energy and mobility systems.

While Europe has committed through the Green Deal to reduce CO₂ emissions by 55%, our usage is exploding (e-commerce, daily mobility, digital services), and the energy mix, along with transportation solutions, is becoming increasingly complex. Faced with the proliferation of various solutions, new energy sources, new vehicles, and new services, the challenge is no longer just about production.

It’s about orchestration: coordinating flows to optimise resource usage without degrading service quality. Given the scale of this task, industry players and public actors are now mobilising AI. Both as a management tool and for optimising systems that have become too complex. This article highlights initiatives and the very concrete needs expressed by industry players to collectively achieve our commitments.

From Corrective to Predictive Maintenance

In the data era, continuously analysing systems and detecting maintenance needs has naturally become standard practice. Predictive maintenance has already proven itself. For example, RAILwAI anticipates rail deformation up to 14 days in advance.

But now, the ambition goes further. It’s no longer just about predicting a shutdown to repair it in time, but about remotely managing systems through data to optimise their performance and lifespan.

This is what Olivier Liandrat, R&D engineer at Reuniwatt, describes. AI enriches cameras and satellite images to predict cloud movement and available light intensity. Result: more reliable photovoltaic production and reduced reliance on carbon-based backup systems like diesel generators.

The challenge is even more dizzying for RTE. The operator must optimise the electrical grid at the country scale, facing increasingly intermittent and decentralised energy production. From statistical models to machine learning approaches, AI has long permeated its analyses. But recent progress now enables direct action in the field:

• Voltage regulation
• Assistants for technicians
• Computer vision to detect defects

However, as Vincent Lefieux from RTE reminds us, this scaling up cannot happen without foundational work on data governance, sharing, and a clear AI adoption strategy.

AI and Mobility: Reversing the Energy Paradigm

For a long time, the electrical system was designed around flexible production: power plants were modulated to follow demand. With the rise of intermittent renewable energies, the paradigm is reversing. Production becomes uncontrollable, and now it’s consumption that must become flexible.

With Smart Charging (V1G), the idea is simple: charge electric vehicles when energy is abundant, green, and cheaper. EDF’s models combine renewable production, prices, grid constraints, and charging habits to adjust power at the right time. Result: up to 30 to 50% energy savings. And as a bonus effect, battery lifespan is extended by several years.

The next step is Vehicle-to-Grid (V2G). The car no longer just consumes; it stores and reinjects energy during consumption peaks. At fleet scale, the vehicle fleet becomes a gigantic distributed battery that can be mobilised to stabilise the grid. To achieve this, actors like LoadStation use AI to predict energy needs for trips based on driver habits, weather, temperature, or traffic, and to orchestrate V2G flows without the user having to worry about it. Experiments are already underway in Gard as part of the Flexitanie project.

AI’s strength is operating behind the scenes. Users continue to plug in “as usual,” while algorithms constantly adjust when to consume, how much to draw from the grid, and on which equipment.

Orchestration: Streamlining Flows Without Building

About 6% fewer vehicles are enough to decongest a network.

Traffic fluidity is primarily a human coordination problem: too many vehicles, in the wrong place, at the wrong time. Hypervisoul offers the AI Flow Controller. This AI agent embedded in connected vehicles recommends speed in real time to smooth traffic and reduce emissions. Experiments have shown that a single connected car can improve traffic flow on highways. The positive impact can extend about 5 kilometres upstream. The next step is direct orchestration between autonomous vehicles.

Freight flows represent 30 to 40% of traffic. And yet, 9 out of 10 transportation plans are still not optimised. This is precisely what Deki is trying to correct by reorganising routes via an AI-powered optimisation engine. For human flows, Transway plays another role. The company encourages responsible mobility choices while collecting valuable data on trips to better understand how we actually move.

Guillaume Desveaux, from AI Cargo Foundation, whose mission is to provide a neutral framework for measuring, sharing, and exploiting flows, points out several obstacles: the heterogeneity of standards and the risk of over-complication. He also reminds us that mobility remains above all a social issue.

Technological Hybridisation: Managing Complexity Through AI

Olivier Flebus, Digital & AI Leader at Schaeffler, presented a very concrete example of this hybridisation between physical models and AI: self-reconfigurable batteries, where each cell is individually controllable.

The objective:

• Continuously adjust charging phases
• Bypass failing cells
• And ultimately, reduce charging time by 20% while increasing battery lifespan by 20%

But managing such a system causes the number of parameters to consider to explode: voltages, temperatures, lifespans, charging times… The physical laws exist, but solving them precisely in real time becomes too costly. This is where AI takes over.

Schaeffler combines reinforcement learning to learn control strategies, surrogate models (ML) to approximate physical models, and generative AI to produce synthetic data. In practice, AI serves to create digital twins to quickly design and test strategies before investing in heavy prototypes, then to embed more frugal models at the very heart of batteries, despite computational constraints.

This example is just one case among others. Solutions are multiplying, sensors too, and we’re accumulating ever more data. With each new service, dozens of additional parameters become adjustable. Thus, our world becomes more complex. And the quest for optimisation through this growth in the number of variables is found in all domains. This complexity can no longer be handled solely with closed and exact models. It requires approximations guided by AI.

Material and Lifecycle: AI in Service of Hardware

With expensive physical assets like batteries, every year of life gained counts for both climate and costs. AI then fits into a logic of making things last rather than throwing them away and replacing them.

Youness Lami explains that within Batconnect, connected lithium batteries (light vehicles, stationary storage) embed a bidirectional tracker: status monitoring, diagnostics, and remote interventions. These batteries generate massive volumes of data that feed an AI chain covering design, operation, and after-sales service. AI enables better understanding of ageing, adjusting usage, and therefore extending lifespan through smarter charging.

It also serves to classify batteries at the end of first use. By quickly estimating their state of health, we decide which ones can go to a second life and which must be discarded. We thus move from simple end-of-life replacement to fine lifecycle management, where each battery is optimised and then redirected.

The Importance of Data: Escaping the Abundance Paradox

If AI is the engine of this transformation, data is its fuel. Yet a paradox persists: we’ve never produced so much data, and yet we exploit very little of it.

Jean-Michel Estibals (RAILwAI) makes a stark observation: the railway sector has never produced so much data. Yet only 15% of this windfall is actually used. The rest sleeps in servers, generating unnecessary storage costs. The challenge is no longer capture, but relevance and architecture.

Should everything be sent to the Cloud? For actors like SIREA or Batconnect, the answer is no. With connected objects generating terabytes of telemetry, the trend is toward Edge Computing: processing information directly on the equipment (the battery, the sensor) to only transmit useful information. It’s a double gain: we reduce latency for security and drastically decrease the digital carbon footprint.

Finally, as Guillaume Desveaux points out, mobility data remains fragmented. Making data from air, road, and maritime transport communicate is a titanic standardisation challenge. Without a common language or regional “Data Places” as envisioned by the MIDOC program, AI will remain myopic, unable to see the entirety of the supply chain.

The Human Factor: Supporting Rather Than Imposing

The most sophisticated technology remains ineffective if it encounters rejection or inertia of habits. AI must not be perceived as a control tool, but as a support lever.

This is the subject raised by Transway and the MIDOC research program. Rather than imposing constraints, AI analyses behaviours. It then proposes personalised alternatives and gently encourages users to change transportation modes. This acceptability is equally crucial in the professional world. Thierry Mourot-Leclercq testifies to SNCF’s strategy: “acculturate rather than replace.” With a goal of zero breakdowns by 2030, the group trains its agents to use generative AI as an assistant, putting humans back at the center of decision-making.

Finally, Hypervisoul reminds us that trust is the keystone of autonomous systems. Should we let an AI control our autonomous car to smooth overall traffic? We will have to accept ceding slightly individual control to gain collective efficiency.

Sovereignty and Geopolitics: Europe Facing Its Destiny

François de Bertier (TENLOG) reminds us that these technical issues fit into a broader geopolitical reality. The recent Draghi report cast a harsh light on the situation. Europe is caught between the urgency to decarbonise and a strong industrial and material dependence on China.

Developing trustworthy AI, capable of optimising our resources without depending on foreign black-box models, becomes a matter of national sovereignty. As the speaker summarises: we cannot be content to be consumers of other countries’ innovations, at the risk of seeing our energy transition piloted from abroad.

The quest for sovereignty, however, clashes with the urgency of climate action. Should we wait for 100% European solutions? Pragmatism requires rapid decarbonization, even if it means accepting temporary technological dependence to rebuild the sector. Moreover, regulation, although perceived as an obstacle, is essential to align the economy and ecology. Without strong regulation, the market alone will not guarantee sobriety. AI must therefore reconcile the acceleration of immediate transition and the construction of future independence.

Occitanie: Full-Scale Laboratory for Trustworthy AI

Faced with these planetary challenges, action crystallises at the territorial scale. Occitanie has chosen not to undergo the technological wave, but to anticipate it with a clear ambition: to become the European leader in critical and trustworthy AI.

The signals are strong: the Region is deploying a massive plan of 60 million euros and structuring its approach through a “Mobility Sector Contract” that brings together industry and academia. But beyond budgets, it’s scientific excellence that constitutes the true engine of this strategy, with a central role devoted to the ANITI institute.

As François-Marie Lesaffre reminded us, AI that stays in the laboratory is useless AI. However, to leave the laboratory and integrate into high-risk industries, such as aeronautics or energy, AI must overcome a major obstacle: certification. Industry players cannot afford to integrate “black boxes” into critical systems.

Hybrid AI

This is where ANITI brings a strategic response: Hybrid AI. The principle is not to oppose data and physics, but to marry them: integrate fundamental laws (mechanics, thermodynamics) into the very heart of learning algorithms. This approach ensures that AI avoids predictions non-compliant with physical laws, compensates for the lack of rare data through theoretical knowledge, and produces more complete and coherent decisions. ANITI doesn’t just do research; the institute builds the trust framework that will allow industry players to move from prototype to industrialisation.

This ecosystem is completed by real-world testing. The MIDOC Key Challenge bridges this cutting-edge research and citizen uses, aiming to create a sovereign “Data Center” for mobility.

Toulouse Métropole, finally, plays the role of a trusted third party to transform the city into an experimentation ground, particularly on the complexity of urban logistics. Occitanie, therefore, doesn’t just host technologies. It builds the complete value chain, from secure mathematical equations to the last mile of delivery.

From Technological Promise to Collective Responsibility

The Future Intelligence day showed it bluntly: the technological building blocks are there. Algorithms know how to predict, optimise, and recommend. Sensors now cover rails, roads, batteries, and electrical grids. Use cases are multiplying, and gains are measurable, immediate, sometimes spectacular.

But one thing is clear: the energy and mobility transition will not be an addition of isolated solutions. It will be systemic, or it will not be.

AI is not the miracle solution to our energy impasses. It is an orchestration tool, capable of transforming rigid systems into adaptive systems, provided it is sober, explainable, governed, and accepted. Without shared data, without common standards, without human trust, it will remain underexploited, or worse, rejected.

Faced with climate urgency, Europe no longer has the luxury of waiting for perfect solutions nor that of renouncing its sovereignty. It must advance on a ridge line: deploy quickly, but intelligently; cooperate without dispossessing itself; regulate without stifling.

Occitanie sketches a possible path: that of hybrid AI, anchored in physics, certifiable, designed for critical systems and tested at the territorial scale. An AI that doesn’t seek to replace, but to augment; not to always build more, but to better use what already exists.

Because fundamentally, the question is not whether AI will transform our mobility and energy systems. It will. The real question is: who will hold the conductor’s baton, and in what direction will the music be played?

Next Edition of Future Intelligence

To have the choice of which score to play, this quest for mastery and protection will continue on March 24, 2026 at ISAE-SUPAERO. Future Intelligence will organise its new edition on AI in the service of Defence and Security, to explore how to build together, in the face of crises, a truly resilient society.