Ruoyu · Jiutian_若愚科技

Self-developed Multi-modal Large Model, MIIT filing

Our self-developed multimodal large model tops the OpenCompass leaderboard, demonstrating superior model capabilities.

The model has obtained MIIT registration and supports domestic Chinese chips and servers, ensuring full autonomous control.

Fusing multimodal data including video, audio, text, point clouds, haptic sensing, etc.

Fine-tuning with a small amount of new data achieves more generalized perception and recognition.

Understanding instructions, environment and self-state, enables real-time dynamic interaction.

Accurately locating affordance regions and recognizing object interactivity.

Decomposing abstract tasks through chain-of-thought reasoning eliminates the need for manual rule engineering

Real-time adaptation of task plans to better cope with the complexities of real-world environments

High-granularity action refinement for fast generation of actionable sequences.

Real-time generation of end-effector trajectories ensures precise execution.

Integrating Ruoyu · Jiutian robot brain and cerebellum, ready to use out of the box.

Providing open edge API services for secondary development.

AI computing power	100 TOPS
GPU	Architecture: 1024-core NVIDIA Ampere GPU with 32 Tensor Cores
GPU	Frequency：918MHz
CPU	Architecture: 8-core Arm® Cortex®-A78AE v8.2 64-bit, 2MB L2 + 4MB L3
CPU	Frequency：918MHz
Memory	16GB 128-bit LPDDR5
HDD	1TB SSD