AISCO - Artificial Intelligence for Solar Car Optimisation - will revolutionize the Design process of Solar Race Cars. AISCO will define new ways of approaching sophisticated Design Challenges and disrupt existing methods.
AISCO aims to help aCentauri in knowledge-transfer, to avoid repeatig past mistakes (LLM) and to accelerate the aerodynamic design process by exploring a large variety of aeroshells and ranking generated aeroshells by minimal racetime (Focus Project).
The assembler bridges engineering constraints and AI-driven design. Our aeroshell must fit within precise spatial boundaries determined by multiple overlapping requirements. The World Solar Challenge Committee mandates specific regulations, e.g. vision standards and wheelbase dimensions that cannot be violated. Physical components like the battery pack and occupant cell further constrain the available design space. The assembler takes these regulations and requirements as inputs, then intelligently positions primitive components while ensuring the AI-generated aeroshell respects every boundary condition, transforming abstract constraints into buildable geometry.
Our vision is to explore a variety of new aeroshell designs. The optimiser maximises performance of aeroshells by minimising a cost function, consisting of multiple weighted losses. Losses range from how well no build zones are respected over aerodynamic performance to solar power performance. Provided that we weight each loss separately, we can generate new aeroshell forms which have never been produced before. Finally, we provide a ranking of aeroshell designs by minimal racetime. This aims to accelerate the exploration process for our aerodynamics team.
High-fidelity RANS simulations are computationally intensive, often taking hours to converge on a single design. This bottleneck makes it impractical to iterate through the many geometries required for optimal aerodynamic performance. To overcome this, we utilize Deep Learning to build surrogate models. Instead of iteratively solving flow equations, our AI predicts the aerodynamic drag nearly instantaneously based on vehicle geometry. The prediction is integrated directly into our aeroshell optimizer as a loss function. This creates a rapid feedback loop where designs are evaluated extremely quickly.
Solar power input depends on factors such as incidence angle, weather, degree of bending, incline angle of the solar deck, solar cell placement, string arrangement and more. Almost all of these factors are dependent on the aeroshell itself, which introduces our solar performance loss: the estimated solar power is fed back into the aeroshell optimiser as a loss function
We want users of our aeroshell optimisation model to be in control of their design. Thus, we are developing a GUI (Graphical User Interface) to change specific constraints and regulations. This will be particularly relevant for the Focus Rollout.
aCentauri has been active for over four years and has successfully competed in multiple races. As a student-led team, member turnover is inevitable: experienced members graduate while new ones join each season. This makes effective knowledge transfer critical to the team’s long-term success.
To address this, we are developing an internal knowledge tool: The aCeBot, a Large Language Model wrapped around our internal documentation, chats, and project history. The goal is to provide fast, structured access to relevant information, helping new and existing members avoid repeating past mistakes and work more efficiently in a high-pace development environment.
The AISCO team consists of 9 ETH Focus Students and 9 Freelancers. The LLM team operates as an independent subsidiary led entirely by Freelancers, while the AISCO Focus Project is driven by Focus Students with support from Freelancers and under supervision of Prof. Fuge’s IDEAL Lab.
