腾讯游戏-多模态大模型算法研究员-动作生成方向
Core
Researching and optimizing multimodal large model architectures specifically for 3D action generation, focusing on the fusion of text, visual, and 3D motion sequences.
Role type
Senior IC multimodal large model researcher (action generation)
Builds
Production multimodal models for character animation, motion capture refinement, and environment-aware interaction.
Domain
AI / Computer Graphics / 3D Animation
Deliverable
production ML models
Required skills
Deep learning theory, Transformer architecture design, End-to-end pretraining, SFT, RL fine-tuning, Python, PyTorch, 3D skeletal animation, Human kinematics (IK/FK, SMPL), Physics simulation (PhysX, Isaac Gym, MuJoCo), Top-tier conference publication experience.
Preferred skills
Motion capture refinement, Reinforcement learning for action control, Open source contributions, Algorithm competition wins.
Technologies
PyTorch, PhysX, Isaac Gym, MuJoCo, SMPL, Transformer
Responsibilities
Design multimodal fusion architectures for text, visual, and 3D action data; Develop tokenization methods for high-dimensional 3D action sequences; Explore scaling laws for multimodal action models; Research multimodal alignment models for voice, text, and action; Develop models for arbitrary character animation generation; Create models for motion capture noise removal; Research environment-aware interactive animation generation.
Seniority
Senior, hands-on IC
