Robotics Foundation Models: Research Scientist & Engineer

Dover; Paris; On-site; EUR 90,000 - 135,000; Full time

Job summary

UMA is seeking a Research Scientist/Engineer to develop architectures and training methods that make robotic foundation models more capable and efficient to train. You will focus on pretraining and explore multimodal architectures, faster training recipes, and compute-efficient scaling strategies for real robots.

You will collaborate with AI researchers and Training Infrastructure engineers to translate architectural changes into scalable training systems and to allocate compute effectively.

Qualifications

• Experience in training large-scale frontier foundation models such as LLMs, Video Generation or Vision Encoders.
• Strong expertise in deep learning research, efficient model training, foundation models, or ML systems.
• Strong Python and PyTorch, with hands-on experience implementing architectures and running substantial training experiments.
• Deep understanding of model architectures, optimization, and training dynamics; able to explain impact on convergence, stability, or compute cost.
• Experience improving training efficiency through architecture, optimization, precision, or hardware-aware implementation.
• Ability to reason about compute, memory bandwidth, tensor shapes, and GPU execution; connect profiling results to research decisions.
• Strong experimental judgment: meaningful baselines, fair compute comparisons, reproducible results, and careful interpretation of scaling trends.
• Interest in connecting efficient training research to multimodal learning and real-world robot capabilities.
• Thrive in a hands-on startup environment; autonomous, rigorous, and collaborative.
• Bonus: Triton, CUDA, fused kernels, efficient attention, or compiler-based optimization.
• Bonus: scaling laws, compute-optimal training, mixture-of-experts models, or large-scale multimodal pretraining.
• Bonus: distributed training, low-precision training, vision-language-action models, or robot learning.
• Bonus: open-source contributions, reproducible research implementations, or technical writing.

Responsibilities

• Research efficient model architectures for robotic foundation models, including attention mechanisms, multimodal fusion, visual and action tokenization, sequence compression, and dense vs sparse or mixture-of-experts designs.
• Optimize training recipes and convergence through optimizers, learning-rate schedules, initialization, normalization, batch sizes, and numerical precision.
• Co-design architectures and compute kernels: identify expensive operations, explore hardware-friendly alternatives, and prototype fused or custom kernels using tools like Triton or CUDA.
• Investigate efficient attention and sequence processing, reducing cost of learning from video, temporal context, and action sequences.
• Develop scaling laws and compute-allocation studies to understand performance vs model size, data, and compute.
• Research training acceleration techniques such as progressive resolution/context, selective freezing, parameter sharing, and lower-precision computation.
• Run controlled ablations and performance benchmarks tracking wall-clock time, compute, memory, convergence, and downstream capability.
• Diagnose training dynamics and numerical instabilities, including gradient behavior and loss spikes as models scale.
• Evaluate how pretraining improvements transfer through mid-training and post-training to real-robot capabilities.
• Work with Training Infrastructure engineers to integrate validated methods into scalable, reproducible training systems, and with Data researchers on data mixtures.

Skills

Python

PyTorch

Deep learning

Research

Experimentation

Hardware-aware optimization

Triton

CUDA

Distributed training

Profiling tools

Tools

Triton

CUDA

Distributed training

Profiling tools

Job description

UMA is seeking a Research Scientist/Engineer to develop architectures and training methods that make robotic foundation models more capable and efficient to train. You will focus on pretraining and explore multimodal architectures, faster training recipes, and compute-efficient scaling strategies for real robots.

You will collaborate with AI researchers and Training Infrastructure engineers to translate architectural changes into scalable training systems and to allocate compute effectively.