Sharpa Robotics
Posted 2mo ago

具身基座模型研究员

Sharpa Robotics
Shanghai, Shanghai, China
OnsiteFull Time
Responsibilities
  • researching models
  • developing models
  • tracking trends
Requirements
  • Bachelor's or higher with research experience in large models
  • Robotics or dexterous manipulation
  • Publications or top-competition awards preferred
  • Experience with vision/3D/video/world models
  • Multimodal representation learning and large-scale training is a plus
Technical tools mentioned
FSDPMegatronvllmdiffusion model

Job description

1. 面向灵巧、精细、长程具身任务,基于海量数据,研发通用、高效、泛化、鲁棒的具身基座模型,效果尽可能地接近人类操作水平。
2. 实时跟踪和关注行业动态和前沿技术发展,保持新技术、新方法的预研和导入。

1. 本科或以上学历,具有大模型、机器人、灵巧手等相关研究经历。
2. 有机器人领域顶会顶刊、CV/LLM 领域顶会文章发表者优先; 顶级学术比赛获奖者优先;至少满足其中至少一条。
3. (加分项)有高引用代表作 and/or 高影响力开源项目。
4. (加分项)有vision-language-action model 或 world action model (夹爪/灵巧手/智驾)相关研究基础,并具备一定的大规模训练经验。
5. (加分项)有video/3D world model 或omni/unified model 相关研究基础,对diffusion model等生成模型基础原理熟练掌握。
6. (加分项)有 视觉/触觉/力觉 等模态的表示学习和生成建模经验。
7. (加分项)有 LLM/diffusion model大规模训练经验,掌握常用的训练和推理引擎(FSDP/Megatron/vllm)。

About Sharpa Robotics

Developing high-performance humanoid robots and core robotic components.

Similar jobs

Embodied Foundation Model Researcher roles near Shanghai, Shanghai
2mo
Save
Mark Applied
Hide
具身基座模型研究员
Shanghai, Shanghai, China
OnsiteFull Time
Sharpa
Sharpa: Building general-purpose humanoid robots and dexterous robotic hands.
Bachelor's degree or higher with research experience in large models, robotics, or dexterous manipulation; publications or top-competition awards preferred; experience with large-scale training and multimodal models; familiarity with FSDP/Megatron/vllm.
FSDP, Megatron, vllm
2mo
Save
Mark Applied
Hide
具身基座模型研究员(实习生)
Shanghai, Shanghai, China
OnsiteInternship
Sharpa Robotics
Sharpa Robotics: Develops high-performance dexterous robotic hands and humanoid platforms.
Bachelor or above in a related field; research experience with large models, robotics or dexterous manipulation; publications or competition awards preferred; familiarity with vision/3D/video models and multimodal representation learning is a plus.
FSDP, Megatron, vllm