About the job
As the Technical Lead Manager (TLM) for the Physical AI team of Scale, you will bridge the gap between cutting-edge Machine Learning research and physical robot deployment. You will lead a high-performing team of Research Engineers while remaining a hands-on technical contributor (~60% of your time). Your primary focus will be the development and evaluation of Large-Scale Foundation Models (e.g VLAs, World models) that allow robots and AVs to generalize across diverse tasks, environments, and morphologies.
Responsibilities
Model Scaling: Direct research into scaling laws for Physical AI, determining how to best utilize massive datasets for pre-training and fine-tuning generalist policies.
VLA and World model development: Develop novel methods for developing and evaluating models, including new Physical AI industry benchmarks
Hands-on Modeling: Actively write code to implement, train and test SOTA architectures. Conduct research on Physical AI data collection, cross-embodiment training, and policy fine-tuning.
Data Strategy: Collaborate with internal labeling teams to design "robotic-native" data pipelines, including the use of VLMs for automated trajectory annotation and data synthesis.
Collaborate closely with customers to drive the industry forward in using Scale data
Mentorship: Lead and grow a team of 4-6 elite Physical AI researchers, fostering a culture of high-velocity experimentation and rigorous evaluation.
Paper-to-Product: Translate the latest research from NeurIPS, ICRA, and CVPR into production-ready features for Scale’s Physical AI partners.
Cross-functional Alignment: Work with cross-functional teams (e.g Product and Operations) to bring our research breakthroughs into production.
Qualifications
Minimum
Deep Learning Mastery: Expert-level proficiency in PyTorch, with deep knowledge of Transformer architectures, Attention mechanisms, and Self-Supervised Learning.
VLM/VLA Experience: Proven track record of working with Vision-Language Models (e.g., CLIP, PaLM-E) and adapting them for spatial reasoning or embodied tasks.
Generative AI: Experience with Diffusion Models for sequence generation or Generative World Models for predictive modeling.
Embodied AI: Strong understanding of Physical AI stack, including imitation learning, reinforcement learning (RL), and multi-modal sensor fusion.
Infrastructure: Experience with large-scale distributed training across GPU clusters and high-performance data loading.
Leadership: 1+ years of experience leading technical teams or projects in a research-intensive environment.
Preferred
Publication Record: First-author publications at top-tier AI/ML conferences (NeurIPS, CVPR, ICRA, CoRL).
Hardware Generalization: Experience building models that work across different robot types (arms, mobile bases, humanoids).
Sim-to-Real: Experience with high-fidelity simulators (e.g., Isaac Gym, MuJoCo) and the nuances of physical domain adaptation.