A
AGIBOT Tech Stack
General-purpose embodied robots with foundation models for industrial automation
AgiBot builds hardware-software integrated robots anchored in a three-pillar AI architecture (manipulation, interaction, locomotion intelligence) and underpinned by proprietary foundation models. The tech stack—LLaVA, LLaMA, Vision Transformer, PyTorch, DeepSpeed, Megatron-LM—reflects heavy investment in large-scale model training and inference at the edge. Hiring velocity is accelerating with five open research roles focused on VLA/VLM pre-training, reinforcement learning post-training, and simulation environments, signaling a company still scaling foundational AI capability rather than pure sales execution.
Tech Stack 12 technologies
What AGIBOT Is Building
◆Challenges
- Compliance risk
- Training efficiency bottlenecks
- Distributed training stability
- Cluster utilization optimization
▲Active Projects
- Vla algorithm development
- Vlm and vla pre-training research
- Large-scale foundation model research
- Generative simulation environment development
- Ai-driven product development
- Dexterous hand manipulation models for embodied intelligence
- Industrial landing of dexterous hand technologies
- Real-robot reinforcement learning post-training of embodied large models
- Improving sample efficiency of real-robot reinforcement learning
- Multi-agent reinforcement learning and distributed real-robot post-training
Hiring Activity
8 roles · 8 in 30d
Department
Seniority
Company intelligence
