Context

As part of a collaboration between Siléane and LAAS-CNRS, we offer a research internship position to contribute to MAGMA-Benchmark, a benchmarking framework for long-horizon, multi-agent robotic manipulation driven by natural language instructions.

MAGMA-Benchmark is a core component of the broader MAGMA research project, which aims to study how learning-based agents can generate, decompose, and execute complex manipulation behaviors over extended time horizons and multiple robots.

The internship will contribute to an ongoing research effort, with the objective of submitting a paper to the International Conference on Robotics and Automation (ICRA 2027): submission September 2026.

Objectives

The intern will work at the intersection of robotics, learning, and software engineering, with a strong focus on Python development. The main objectives include:

• Methodological Innovation: Propose and explore training strategies for coordinating multiple robots using language-conditioned policies.
• Benchmark Architecture: Design and implement modular, long-horizon tasks in Python, contributing to the MAGMA-Benchmark ecosystem.
• Scientific Contribution: Participate in defining evaluation metrics and co-author a submission to ICRA 2027.
• Tooling & Pipeline: Improve the MAGMA-GEN pipeline (abstractions, training setups), applying clean code principles to cutting-edge research.

While the internship involves substantial implementation work, it is not limited to executing predefined tasks. The intern is encouraged to take initiative, experiment with ideas, and contribute creatively to both the benchmark and the learning pipeline.

Expected skills

We are primarily looking for a strong Python developer with an interest in learning-based systems.

Required

• Good proficiency in Python
• Experience with PyTorch
• Strong software engineering skills (clean code, modular design, experimentation)

Highly appreciated

• Previous Python projects (academic or personal)
• Knowledge of Git
• Experience with machine learning or reinforcement learning
• Familiarity with robotics concepts or simulation environments

Optional

• Experience with Gym / Gymnasium-like environments
• Robotics background (ROS, simulation, manipulation)

Students with a machine learning or computer science background are strongly encouraged to apply, even without deep robotics expertise.

Practical information

Duration: 4-6 months.

Location: LAAS-CNRS Toulouse - GEPETTO Team

Start-Date: March 30 (flexible)

Funding: Yes

Langage: English or French