Futures
Access hundreds of perpetual contracts
TradFi
Gold
One platform for global traditional assets
Options
Hot
Trade European-style vanilla options
Unified Account
Maximize your capital efficiency
Demo Trading
Introduction to Futures Trading
Learn the basics of futures trading
Futures Events
Join events to earn rewards
Demo Trading
Use virtual funds to practice risk-free trading
Launch
CandyDrop
Collect candies to earn airdrops
Launchpool
Quick staking, earn potential new tokens
HODLer Airdrop
Hold GT and get massive airdrops for free
Launchpad
Be early to the next big token project
Alpha Points
Trade on-chain assets and earn airdrops
Futures Points
Earn futures points and claim airdrop rewards
Gaode announces the full open-source of the embodied operation base model ABot-M0, enabling a "general brain" to adapt to various forms of embodied robots.
Sina Technology reported on April 1 in the afternoon that, recently, Amap announced the official full-scale open-source release of ABot-M0—the world’s first embodied operation foundation model built on a unified architecture—which can enable a “universal brain” to adapt to embodied robots in multiple forms.
据 reports, the model has achieved SOTA performance across multiple authoritative benchmark tests, including Libero, Libero-Plus, RoboCasa. In the Libero-Plus benchmark, the model’s task success rate reached 80.5%, improving by nearly 30% over the industry’s previous benchmark solution Pi0.
This open-source release of ABot-M0 covers three major dimensions: data, algorithms, and the model. It aims to break through data silos and deployment challenges, providing cutting-edge spatial understanding capabilities for the embodied intelligence field, as well as a “plug-and-play” universal technical foundation.
On the data layer, ABot-M0 opens UniACT, currently the largest-scale general-purpose robot dataset, integrating more than 6 million real-world operation trajectories, and providing a full end-to-end processing pipeline from raw heterogeneous data to standardized training data. On the algorithm layer, ABot-M0 simultaneously opens the model architecture and training framework, including the Action Manifold Learning (AML) algorithm and the dual-stream perception architecture proposed by Amap Innovation, as well as the dual-stream perception architecture. On the model layer, ABot-M0 also open-sources end-to-end pretrained models and a complete toolchain, so developers can quickly adapt to scenarios such as industrial and home settings without having to build a training framework from scratch.
“We believe that truly universal embodied intelligence is not the product of working in isolation behind closed doors, but the result of global developers jointly refining it.” The technical lead for Amap’s ABot-M0 said, “We look forward to ABot-M0 becoming a bridge connecting academic research and industrial applications, so that every robot can have a smart, reliable, universal ‘brain’.”
Massive information, precise analysis—exclusively on the Sina Finance app
责任编辑:张乔松