Embodied VLA Algorithm Engineer | VinDynamics

Proficient in Transformer architecture, with training and tuning capabilities for VLA/VLM/LLM and other models, and familiar with the principles and applications of mainstream models such as RT-2, OpenVLA, PI, OpenEMMA, and EMMA

Master multimodal alignment technology to realize the end-to-end generation of vision and language input to the robot's action sequence (such as task planning and action control)

Proficient in using PyTorch, DeepSpeed and other frameworks, and has experience in distributed training of large models with multiple machines and multiple cards

Optimize the inference efficiency of models on embedded platforms (e.g., lightweight deployment, CUDA acceleration) to support real-time control of real robots or autonomous driving systems

Combined with imitation learning (IL) and reinforcement learning (RL), the generalization of the model in physical scenarios is enhanced