Inspired by biological growth, we propose SMOL to dynamically modulate robot strength during learning. This curriculum helps agents leverage advantageous physics early on, improving both performance and behavior diversity.
@inproceedings{templier2025time,title={Time to Play: Simulating Early-Life Animal Dynamics Enhances Robotics Locomotion Discovery},author={Templier, Paul and Janmohamed, Hannah and Labonte, David and Cully, Antoine},booktitle={Artificial Life Conference Proceedings 37},volume={2025},number={1},pages={71},year={2025},month=sep,organization={MIT Press One Rogers Street, Cambridge, MA 02142-1209, USA journals-info~…},}
AURORA-XCon extends unsupervised Quality-Diversity (QD) to solve deceptive optimization problems without expert-defined descriptors, even out-performing traditional QD methods with hand-crafted features.
@inproceedings{coiffard2025overcoming,title={Overcoming Deceptiveness in Fitness Optimization with Unsupervised Quality-Diversity},author={Coiffard, Lisa and Templier, Paul and Cully, Antoine},booktitle={Proceedings of the Genetic and Evolutionary Computation Conference},pages={122--130},year={2025},month=jul,}
@inproceedings{cully2025evolutionary,title={Evolutionary Reinforcement Learning},author={Cully, Antoine and Lim, Bryan and Templier, Paul and Flageat, Manon},booktitle={Proceedings of the Genetic and Evolutionary Computation Conference Companion},pages={1095--1124},year={2025},month=jul,}
@article{templier2025leveraging,title={Leveraging Structures in Evolutionary Neural Policy Search},author={Templier, Paul},journal={ACM SIGEVOlution},volume={18},number={1},pages={1--3},year={2025},publisher={ACM New York, NY, USA},}
The Geometric Encoding for Neural Network Evolution (GENE) relies on pseudo-distance functions to encode neural networks as smaller genomes. Using Genetic Programming as a meta-evolution loop, we learned a new encoding based on GENE. The discovered encoding makes sparse networks emerge naturally.
@inproceedings{kunze2024searching,title={Searching Search Spaces: Meta-evolving a Geometric Encoding for Neural Networks},author={Kunze, Tarek and Templier, Paul and Wilson, Dennis G},booktitle={2024 IEEE Congress on Evolutionary Computation (CEC)},pages={1--8},year={2024},month=jul,organization={IEEE},}
Best Paper Award in the Complex Systems Track at GECCO 2024
Quality with Just Enough Diversity (JEDi) uses behavior information from Quality-Diversity to improve the search capabilities of ES, by learning and focusing on interesting behaviors. (Follow-up work under review).
@inproceedings{templier2024quality,title={Quality with Just Enough Diversity in Evolutionary Policy Search},author={Templier, Paul and Grillotti, Luca and Rachelson, Emmanuel and Wilson, Dennis and Cully, Antoine},booktitle={Proceedings of the Genetic and Evolutionary Computation Conference},pages={105--113},year={2024},month=jul,}
By studying the injection of an RL actor into an ES population, we show they drift apart genetically, which can lead to the ES breaking. We introduce GDR, a simple regularization in the actor training loss, to fix it.
@inproceedings{templier2024genetic,title={Genetic Drift Regularization: On Preventing Actor Injection from Breaking Evolution Strategies},author={Templier, Paul and Rachelson, Emmanuel and Cully, Antoine and Wilson, Dennis G},booktitle={2024 IEEE Congress on Evolutionary Computation (CEC)},pages={1--9},year={2024},month=jul,organization={IEEE},}
While training an artificial agent for complex tasks like driving a car, mastering a video game, or controlling plasma in a nuclear fusion reactor, innovations can lead to intelligent behavior. In such scenarios, a promising approach is to mimick the natural world’s evolutionary process, which has honed the problem-solving capabilities of animal brains. Evolutionary Neural Policy Search (ENPS) draws inspiration from this concept. It creates a diverse population of “brains” represented by neural networks, allowing the system to “evolve” by selectively combining and mutating successful individuals. This thesis delves into the core components of ENPS and their intricate interplay. By analyzing the structures of ENPS, the goal is to design novel policy search methods that enhance these components, ultimately leading to the development of more efficient and effective learning algorithms for complex tasks.
@phdthesis{templier2024synergies,title={Leveraging Structures in Evolutionary Neural Policy Search},author={Templier, Paul},year={2024},month=apr,school={Toulouse, ISAE},}
To tackle the impact of uncertain evaluations in genetic algorithms, we introduce LUCIE, a resampling scheme based on a bandit approach. LUCIE is able to better select elite individuals, making the GA more robust to noise. (*equal contribution)
@inproceedings{lecarpentier2022lucie,title={LUCIE: an evaluation and selection method for stochastic problems},author={Lecarpentier, Erwan and Templier, Paul and Rachelson, Emmanuel and Wilson, Dennis G},booktitle={Proceedings of the Genetic and Evolutionary Computation Conference},pages={730--738},year={2022},month=jun,}
Directly optimizing the weights of a neural network with evolution can get expensive, especially with methods like XNES or CMAES. We introduce GENE, a new method to encode neural networks as genomes which keeps the performance of direct encoding while reducing the size of the genome by an order of magnitude.
@inproceedings{templier2021geometric,title={A geometric encoding for neural network evolution},author={Templier, Paul and Rachelson, Emmanuel and Wilson, Dennis G},booktitle={Proceedings of the Genetic and Evolutionary Computation Conference},pages={919--927},year={2021},month=jun,}
GECCO 2020 competition on evolving a Dota 2 bot, with a CGP and NEAT implementation, Quality-Diversity for exploration and a neural-based world model.
@article{templierevolving,title={Evolving a Dota 2 bot: Illuminating search in CGP and NEAT},author={Templier, Paul and Hervier, Lucas and Wilson, Dennis},year={2020},month=jun,}
