Amap, Alibaba’s location-based services platform, has recently unveiled a comprehensive upgrade to its ABot embodied AI system. The upgraded system includes ABot-N1, ABot-M0.5, ABot-ER, ABot-AgentOS, and ABot-C0, addressing critical challenges in embodied intelligence. This technology enables robots to perceive, reason, and act in physical environments effectively.
The enhanced system focuses on enhancing navigation, manipulation, task reasoning, long-term memory, and motion control. Amap claims that the upgraded system has achieved state-of-the-art results on 17 widely used benchmarks. The ABot architecture is now structured as an innovative full-stack embodied-intelligence technology system, connecting world models, foundation models, and an embodied-agent framework. This setup allows data from simulation training, physical interaction, and memory scheduling to enhance system performance over time.
ABot-N1, a general navigation foundation model, is designed to assist robots in moving through open environments. Unlike conventional navigation maps built for human users, ABot-N1 provides robots with finer-grained spatial perception necessary for safe navigation in the physical world. The model uses a dual-system architecture for accurate navigation and introduces a pixel-level chain-of-thought process to improve decision-making.
ABot-M0.5, a general manipulation foundation model, coordinates robot movement and object interaction for effective task completion in real environments. It separates locomotion and manipulation into two action streams and utilizes a training method called “dream self-healing” to reduce failures caused by deviations. In testing, ABot-M0.5 outperformed previous models on complex and basic tasks.
ABot-ER and ABot-AgentOS form the system’s embodied-agent layer, supporting decision-making and turning planning into executable actions. ABot-C0, the motion-control component, translates decisions from the ABot system into physical actions, promoting collaboration among different robot types.
The updated ABot system aims to enable robots to learn and adapt in real-world open environments continuously. Amap has also released research papers detailing the advancements in ABot-N1, ABot-M0.5, ABot-AgentOS, and ABot-C0 for further reference.
About the Author
