About the job
About the Role
In the realm of machine learning, pretraining lays the foundation for a general model, while post-training refines that model, enhancing its utility, controllability, safety, and performance in real-world applications. As a Post-Training Research Scientist, you will transform large pretrained robot models into production-ready systems through methodologies such as fine-tuning, reinforcement learning, steering, human feedback, task specialization, evaluation, and on-robot validation at scale. This position offers a unique opportunity for individuals from diverse backgrounds to evolve into full-stack ML roboticists, adept at swiftly identifying challenges across machine learning and control domains. This is where innovative research converges with practical implementation.
Your Responsibilities Include:
Crafting fine-tuning and adaptation strategies tailored for specific robotic tasks and embodiments.
Developing methodologies to enhance reliability, robustness, and controllability of robotic systems.
Establishing evaluation frameworks to assess real-world robot performance beyond just offline metrics.
Collaborating with ML infrastructure teams to optimize inference-time performance, including latency, stability, and memory usage.
Utilizing advanced techniques such as imitation learning, reinforcement learning, distillation, synthetic data, and curriculum learning.
Bridging the gap between model outputs and tangible outcomes in the physical world.
You Might Excel in This Role If You:
Possess experience in fine-tuning large models for downstream applications, including RLHF, imitation learning, reinforcement learning, distillation, and domain adaptation.
Have a background in embodied AI, robotics, or real-world machine learning systems.
Demonstrate a strong commitment to evaluation, benchmarking, and failure analysis.
Are comfortable troubleshooting and debugging across the entire ML stack, from analyzing loss curves to understanding robot behavior.
Enjoy rapid iteration and thrive on real-world feedback loops.
Aspire to connect foundational models with practical deployment scenarios.
About Generalist
At Generalist, we are dedicated to realizing the vision of general-purpose robots. We envision a future where industries and homes benefit from collaborative interactions between humans and machines, enabling us to achieve more than ever before. Our focus is on building embodied foundation models, starting with dexterity, and advancing the frontiers of data, models, and hardware to empower robots to intelligently engage with their environments.
