Jon Goikoetxea

Pronunciation: [ yon goh-ee-koh-eh-CHEH-ah ]

I'm an Artificial Intelligence MSc student at Ensimag - INP Grenoble.

I'm broadly interested in optimal and learning-based control, with special emphasis on robot learning. My research aims to make autonomous agents more generally capable and efficient by leveraging optimization and machine learning methods.

As an MSc student in Ensimag, I worked on multi-objective reinforcement learning with formal specifications with Prof. Alexandre Donzé in the Verimag laboratory. Before coming to Grenoble, I obtained my BSc at the Public University of Navarre, where I worked with Prof. Jesús Palacián on learning for control in robotics and astrodynamics. I also worked with the UpnaLab research team led by Prof. Asier Marzo, mostly on simulations for acoustic holography.

Check out my projects on GitHub, visit my LinkedIn profile, or email me here.

Publications:

For an up-to-date list, visit my Google Scholar.

Animated GIF of GCImOpt policies controlling various dynamical systems

GCImOpt: Learning efficient goal-conditioned policies by imitating optimal trajectories

Jon Goikoetxea, Jesús F. Palacián
Learning for Dynamics and Control (L4DC) 2026
Webpage - PDF

Projects:

See more on my GitHub page.

Animation of the cart-pole swing up problem and the solution generated by the trajectory optimizer

Trajectory optimizer using direct collocation

Trajectory optimizer based on direct collocation, using Jax for automatic differentiation and the IPOPT nonlinear programming solver. Supports trapezoidal / Hermite-Simpson collocation and iterative mesh refinement.

Animation of a trained Play-LMP policy controlling a Franka robot arm in a tabletop environment, executing different tasks in sequence

Learning latent plans from play

Implementation of the Play-GCBC and Play-LMP self-supervised policy learning methods by Lynch et al. (2019). Trained multi-task robot manipulation policies on dataset from the CALVIN benchmark (Mees et al., 2022).

Mini-GPT

Implementation of a small GPT-like Transformer language model from scratch, from tokenization to model architecture, training and inference. Trained models on the SimpleStories text dataset (Finke et al., 2025).

Screen recording of Game Boy emulator running neural network inference. A six is drawn on the screen, the screen flashes for some seconds and six is marked as the predicted class

gb-nn: Neural network inference on a Game Boy

Implementation of quantized neural network inference on the Game Boy, written by hand in assembly language. The trained NN running on an emulated Game Boy obtains 93.86% accuracy on the MNIST test set.

Illustration of trajectory of a simulated spacecraft controlled by a deep neural network

Adaptive guidance and control for spacecraft orbit transfers using deep reinforcement learning

Neural satellite controller for low-thrust orbit transfers using deep reinforcement learning. Presented as an oral presentation at the XXII Workshop on Celestial Mechanics in Girona, Spain.