AI Research Scientist at Autodesk in Toronto, ON

Job description

Job Requisition ID # 26WD101116 L'affichage de poste en français suivra / The French job posting follows. 26WD101116, AI Research Scientist Position Overview As a Senior Research Scientist in the AI Lab at Autodesk Research, you will conduct fundamental and applied research on Motion Foundation Models: general-purpose models that can represent, generate, edit, understand, and reason about motion. The work will initially focus on human and object motion and will extend toward multimodal and interactive settings involving text, trajectories, images, video, geometry, robot observations/actions, and other signals. You will work with scientists, researchers, developers, and designers across generative AI, computer vision, graphics, robotics, simulation, manufacturing, architecture, and construction to develop new model architectures, representations, training methods, datasets, and evaluations for motion intelligence. Autodesk's AI Lab is active in the wider research community, targeting publications at CVPR, NeurIPS, ICML, ICLR, SIGGRAPH, and other top-tier conferences. We collaborate with leading academic and industry labs, combining the best of an academic research environment with product-guided research. This role will report to a Manager of Research Science in the AI Lab. Responsibilities Develop new foundation-model approaches for motion generation, editing, understanding, reasoning, and representation Drive research end-to-end: identify high-value questions, formulate hypotheses, implement ideas, run experiments, analyze failures, iterate, and communicate conclusions Investigate motion representations and tokenization, generative modeling, multimodal conditioning and alignment, scalable training, post-training, and model refinement Design rigorous evaluations, benchmarks, ablations, and diagnostics that reveal genuine model progress and failure modes Build and modify research code, training pipelines, and data pipelines, using modern AI coding agents where useful while maintaining ownership of correctness, reproducibility, and scientific quality Work toward ambitious long-term goals while defining concrete milestones and maintaining a high pace of empirical iteration Collaborate across a global research team and with academic and industry partners; publish at top-tier conferences and help transfer successful ideas into Autodesk technologies Minimum Qualifications A Master's or PhD in Computer Science, Machine Learning, Computer Vision, Computer Graphics, Robotics, Mathematics, or a related technical field A consistent track record of research publications demonstrating experience in applied or fundamental research Deep knowledge of modern generative or foundation-model methods, such as Transformers, diffusion/flow models, latent-variable models, or related architectures Strong hands-on ability with PyTorch, JAX, or similar frameworks, including building, modifying, debugging, and scaling research training pipelines; effective use of modern AI-assisted development and coding agents Excellent experimental judgment: ability to turn ambiguous research questions into testable hypotheses, controlled experiments, useful metrics, and clear conclusions Demonstrated research ownership, persistence, and execution: a track record of pushing difficult technical problems through repeated experimentation to meaningful results Preferred Qualifications Research experience in human motion, character animation, motion capture, embodied agents, robotics, or other spatiotemporal modeling problems Experience with multimodal or foundation models combining motion with text, images, video, geometry, trajectories, actions, or other modalities Experience with learned motion representations or tokenization (for example VQ/VAE-style methods) and generative methods such as diffusion, flow matching, or autoregressive modeling Research experience with large-scale video foundation models, including video generation, spatiotemporal representations, multimod