Embodied intelligent algorithm: (VLA model, imitation learning, multimodal basic model)
Duties:
1. Responsible for developing a multimodal decision model for a two-arm robot to complete complex physical tasks in a real environment, and pushing the VLA (visual-language-action) foundation model of robots to the ground;
2. Responsible for data collection design, algorithm architecture design, model training, engineering deployment and continuous performance optimization, deep collaboration with hardware, data platform, application team to build end-to-end system-level solutions;
3. Continuously explore innovative applications of multimodal large models (such as VLM, VLA, VLN, etc.) in the embodied intelligence field, and promote algorithmic technology transformation from laboratory to actual scenes;
4. Follow industry and academic frontiers, reproduce, optimize and transform the latest embodied intelligence-related technical achievements.
Requirements:
1. Master's degree or above, majoring in Computer Science, Artificial Intelligence, Robotics, Automation, etc., solidly mastering core algorithms such as robot learning, deep learning, and imitation learning;
2. Proficient in Python, with good C++/Python programming habits, and familiar with mainstream deep learning frameworks such as PyTorch/TensorFlow;
3. Have multi-modal large models (VLM/VLA, etc.) in robot perception, operation, navigation, etc. Research and landing experience, or have in-depth research background on related algorithms;
4. Familiar with visual, language, etc. multi-modal fusion algorithm, with multi-modal large model tuning experience, and understanding of embodied intelligence field frontier innovation progress;
5. Have good self-motivation, stress resistance, and team collaboration, cross-department communication skills, and be passionate about robots' embodied intelligence.
1. Have published influential work in professional conference journals in related fields such as robotics, machine learning, computer vision (TRO, RSS, ICRA, CoRL , NeurIPS, ICLR, CVPR, etc.), or have professional academic competition winning experience;
2. Have large-scale data collection-algorithm co-ordination (data flywheel) system development experience;
3. Familiar with mainstream robot simulation platforms such as IsaacGym/Sim, SAPIEN, MuJoCo, Gazebo; 4. Have relevant experience in the multi-modal large model field or core contributions to open source projects;
5. Familiar with reinforcement learning, robot control algorithms, multimodal perception, etc. related algorithms.
