具身智能大模型算法专家岗位职责:
大模型框架研发:
•研究并构建基于多模态大模型(如视觉、语言、触觉等)的具身智能算法框架,实现机器人在开放环境中的感知、推理与决策闭环。
•探索大模型(如LLM、VLM、VLA、世界模型)与机器人具身学习的结合,提升机器人在物理交互中的泛化能力和适应性。
算法优化与部署:
•设计高效的模型结构(如轻量化、分布式训练、边缘计算),解决具身任务中的实时性、计算资源限制等问题。
•优化强化学习、模仿学习等具身算法与大模型的协同机制,提升机器人复杂任务(如抓取、导航、人机协作)的表现。
开放世界交互能力:
•开发面向开放环境的动态感知与决策系统,使机器人能够处理未知物体、场景变化及长周期任务。
•结合仿真(如Isaac Gym)与真实机器人平台(如机械臂、移动机器人)进行算法验证与迭代。
技术前瞻与落地:
•跟踪学术界与工业界最新进展,推动技术突破与产品化落地。
任职要求:
教育背景:
1.计算机科学、人工智能、机器人学、自动化等相关领域博士学历,或硕士学历+3年以上相关工作经验;
大模型深度经验:
2.3年以上大模型(LLM/VLM/VLA)研发经验,熟悉主流架构(AR、Diffusion、Mixture of Experts)及训练方法(RLHF、SFT);
3.深入理解强化学习基本原理、核心算法(如DQN、PPO、SAC等)及常用框架(如OpenAI Gym、Rlib等),具备将强化学习应用于机器人领域的实际项目经验,如基于强化学习的机器人控制、导航、操作等;
4.有参与大模型全流程开发(预训练、微调、部署)的实际项目,或在顶会(CoRL、ICRA、NeurIPS、ICLR等)发表过具身智能或大模型相关论文;
具身智能技术结合:
5.熟悉机器人学习算法,了解机器人技术框架;
6.了解VLA技术框架, 有Octo,OpenVLA, Pi0等复现经验者优先。
Embodied Intelligence Large Model Algorithm Specialist Responsibilities:
Large Model Framework R&D:
• Research and build embodied intelligence algorithm frameworks based on multimodal large models (e.g., vision, language, haptics) to achieve closed-loop perception, reasoning, and decision-making for robots in open environments.
• Explore integration between large models (e.g., LLM, VLM, VLA, world models) and robotic embodied learning to enhance robots' generalization capabilities and adaptability in physical interactions.
Algorithm Optimization and Deployment:
• Design efficient model architectures (e.g., lightweight structures, distributed training, edge computing) to address real-time constraints and computational resource limitations in embodied tasks.
• Optimize synergistic mechanisms between embodied algorithms (e.g., reinforcement learning, imitation learning) and large models to improve robotic performance in complex tasks (e.g., grasping, navigation, human-robot collaboration).
Open-World Interaction Capabilities:
• Develop dynamic perception and decision systems for open environments, enabling robots to handle unknown objects, scene changes, and long-duration tasks.
• Validate and iterate algorithms using simulations (e.g., Isaac Gym) and real robot platforms (e.g., robotic arms, mobile robots).
Technology Forecasting and Implementation:
• Track cutting-edge advancements in academia and industry to drive technological breakthroughs and productization.
Qualifications:
Education:
- Ph.D. in Computer Science, Artificial Intelligence, Robotics, Automation, or related fields; or Master's degree + 3+ years of relevant experience.
Deep Learning Experience:
3+ years of R&D experience with large language models (LLM/VLM/VLA), proficiency in mainstream architectures (Attention-based, Diffusion, Mixture of Experts) and training methods (RLHF, SFT).
Deep understanding of reinforcement learning fundamentals, core algorithms (e.g., DQN, PPO, SAC), and common frameworks (e.g., OpenAI Gym, Rlib), with practical project experience applying reinforcement learning to robotics domains such as robot control, navigation, and manipulation;
Practical experience in full-cycle large model development (pre-training, fine-tuning, deployment) or publication of embodied intelligence/large model-related papers in top conferences (CoRL, ICRA, NeurIPS, ICLR, etc.);
Embodied Intelligence Integration:
Familiarity with robot learning algorithms and robotics technology frameworks;
Understanding of VLA technology frameworks; experience with Octo, OpenVLA, Pi0, or similar systems preferred.