ACERobotics/ACE-Brain-0.5-8B
ACE-Brain-0.5-8B is an 8 billion parameter unified embodied foundation model developed by the ACE-Brain-Team, designed for Physical Agentic AI. It integrates perception, planning, action, monitoring, and self-improvement into a closed-loop system for robots. This model excels at spatial reasoning, object grounding, long-horizon task planning, and navigation/manipulation action generation within physical environments, supporting a 32768 token context length.
Loading preview...
Overview
ACE-Brain-0.5 is a unified embodied foundation model, extending ACE-Brain-0 into a closed-loop system for Physical Agentic AI. Developed by the ACE-Brain-Team, this 8 billion parameter model enables robots to perceive the physical world, plan towards goals, execute actions, monitor progress, and learn from experience. It integrates five core cognitive functions: Spatial Perception, Decision Making, Embodied Interaction, Self Monitoring, and Self Improvement.
Key Capabilities
- Unified Embodied Intelligence: Organizes robot intelligence into a single closed-loop model for comprehensive physical interaction.
- Perception-Planning-Action Loop: Supports object and affordance grounding, 3D and egocentric spatial reasoning, long-horizon task planning, and navigation/manipulation action generation.
- Execution Monitoring: Includes progress estimation for verification and recovery during task execution.
- SSR+ Training Paradigm: Utilizes an extended Scaffold-Specialize-Reconcile with a Reactivate stage, combining task-vector merging and targeted fine-tuning to unify diverse robotic tasks without interference.
Good For
- Developing advanced physical agentic AI systems.
- Applications requiring complex robot perception, planning, and manipulation in real-world environments.
- Research into embodied AI and closed-loop robotic intelligence.