Open Role
Research Scientist: Post-Training
Singapore / Munich, Germany
Develop end-to-end robot manipulation models that understand tasks, adapt to environments, and operate in the real world.
About the Role
MindOn is building general-purpose intelligence for robots. We are looking for a Research Scientist focused on post-training to train, optimize, and deploy robot manipulation models on real hardware.
You will work on Vision-Language-Action (VLA) models and related robot learning methods, using robot data and human demonstrations to improve generalization across tasks, objects, and environments and bring algorithms into real-world applications.
Responsibilities
- Develop end-to-end robot manipulation models to improve performance on complex manipulation tasks.
- Train, fine-tune, evaluate, optimize, and deploy embodied AI models on real robots.
- Use UMI and related data collection systems and human demonstrations for manipulation learning, addressing the transfer from demonstration data to robot actions.
- Analyze, clean, and organize robot datasets to improve data quality, model performance, and generalization.
- Build training and evaluation workflows and use failure analysis to drive a closed loop of data, training, evaluation, and deployment.
- Collaborate with robot control, data, and infrastructure teams to resolve deployment and execution challenges in real environments.
Qualifications
- A deep and genuine passion for Embodied AI is essential.
- Proficiency with AI coding tools is essential: use them effectively for development, debugging, testing, and troubleshooting; independently review generated code and validate its correctness and reliability through tests and real-world execution.
- A practical understanding of Embodied AI, including its strengths, limitations, and current capability boundaries, and the ability to assess methods against real robot tasks.
- A master’s degree or above in Computer Science, Robotics, AI, Automation, or a related field, or equivalent practical experience.
- Proficiency in Python and familiarity with frameworks such as PyTorch, TensorFlow, or JAX, with experience training, debugging, and optimizing models.
- Knowledge of Diffusion Policy, OpenVLA, π0, In-Context Learning (ICL), and related models and methods, including their principles, applications, and limitations.
- A strong understanding of UMI data collection and manipulation learning workflows, including data acquisition, processing, action representations, and using human demonstrations to train robot manipulation models.
- Practical experience developing robot manipulation models, independently running experiments, evaluating performance, and analyzing failures.
- Strong problem-solving skills, independent execution, and cross-functional collaboration.
Preferred Experience
- Experience deploying and debugging manipulation models on real robots.
- Experience developing models for dual-arm coordination, mobile manipulation, or contact-rich tasks.
- Familiarity with applications of imitation learning, reinforcement learning, world models, and multimodal foundation models.
- Experience with large-scale data processing, distributed training, or inference optimization.
- Relevant open-source contributions or publications at CoRL, RSS, ICRA, IROS, NeurIPS, ICML, ICLR, or similar venues.