Position Overview
We are seeking an Applied Scientist (Robotics Simulation) to design and build the physics-solving engine behind Collinear’s synthetic data generation platform. In this role, you will architect and optimize a next-generation simulation stack engineered purely for maximum data throughput, procedural environmental scale, and physical fidelity.
You will collaborate closely with GPU optimization engineers, procedural scene generation teams, and AI researchers pushing the limits of model training on synthetic data. The ideal candidate brings deep physics simulation expertise and hands-on experience implementing high-performance solvers on modern parallel GPU architectures.
Key Responsibilities
Physics Solver Innovation: Improve and develop advanced physics solvers and numerical modeling methods for high-degree-of-freedom interactions, complex contact dynamics, deformables, and fluids.
Training Data Optimization: Partner with team members to design simulation environments, domain randomization strategies, and physics parameter ranges that yield high-utility training datasets for downstream AI models.
Procedural Scene & Scale Integration: Collaborate with scene-generation engineers to scale environment variations, physical property distributions, and dynamic multi-agent interactions across billions of simulation steps.
Performance Profiling: Profile and optimize simulation execution at scale, minimizing latency and memory overhead in multi-GPU, parallelized rollout pipelines.
Technical Leadership: Help shape Collinear’s long-term roadmap for high-fidelity physical modeling, differentiable simulation, and synthetic data engine architecture.
Preferred Qualifications
Education: MS or PhD in Physics, Computer Science, Applied Math, Engineering, or equivalent hands-on industry experience in physics-based simulation.
Physics Engine Expertise: Strong track record of developing or extending high-fidelity physics engines, rigid-body dynamics solvers, or constraint-based systems. Experience with deformables, fluids, soft bodies, or differentiable simulation is a plus.
Numerical Methods & Dynamics: Deep understanding of classical mechanics, contact modeling, constraint solvers, integrators, and managing accuracy-vs-speed trade-offs for large-scale computation.
GPU Programming: Advanced expertise in CUDA and GPU optimization, with a proven ability to accelerate numerical algorithms on parallel hardware architectures.
Systems Engineering: High proficiency in C++ and Python, with a track record of building reliable, high-throughput software used in production data or ML pipelines.
ML Infrastructure Familiarity: Strong grasp of how ML frameworks consume simulation outputs (e.g., vectorized environments, massively parallel rollouts, synthetic dataset generation).
Fidelity Intuition: Deep intuition for physical realism and dataset distribution drift—understanding how synthetic physical data behaves and how to model environmental variance without relying on physical hardware collection.
Track Record: Publications, open-source contributions, or shipped commercial systems in numerical computing, graphics, physics simulation, or synthetic data generation.