AGIBOT introduces Genie Sim 3.0 simulation platform for embodied AI

Whereas latest progress in robotics has been pushed by advances in fashions and algorithms, real-world deployment continues to be constrained by excessive information assortment prices, restricted state of affairs variety, and fragmented benchmarking requirements, in accordance with AGIBOT. The corporate in the present day stated it has upgraded its Genie Sim 3.0 growth atmosphere.

AGIBOT stated its platform now addresses three long-standing bottlenecks in embodied AI: atmosphere era, information scalability, and standardized analysis. The Shanghai-based firm  stated it designed Genie Sim 3.0 to combine scene era, simulation, information, and analysis right into a unified, reusable infrastructure.

Genie Sim World generates environments from language

Genie Sim 3.0 introduces a spatial world mannequin that permits customers to generate absolutely interactive 3D environments from easy textual content or picture inputs. AGIBOT stated its key capabilities embrace:

• Multimodal enter – No guide modeling or {hardware} setup required. Customers can generate numerous environments with minimal enter.
• Minute-level scene creation – Neural community inference allows scene era in minutes, in contrast with hours in conventional pipelines.
• Excessive-fidelity – Synchronized output of RGB, depth, lidar, and different multimodal information ensures alignment with actual robotic notion

Editor’s be aware: On the 2026 Robotics Summit & Expo on Could 27 and 28 in Boston, there will probably be classes on embodied and bodily AI, in addition to on humanoid robotic growth. Registration is now open.

Genie Sim 3.0 benchmark presents complete analysis framework

For the 5 core capabilities of robotic algorithms—instruction understanding, spatial reasoning, atomic talent operation, disturbance adaptation, and training-to-deployment generalization — AGIBOT stated it has designed 5 corresponding activity suites. Genie Sim Benchmark helps mainstream fashions such because the GO-2, Pi sequence, and GR00T sequence and offers a multi-dimensional, systematic analysis of the fashions’ complete efficiency in complicated situations.

The framework evaluates 5 core capabilities of embodied AI techniques:

• Instruction following (GenieSim-Instruction) – Measures alignment between pure language directions and robotic conduct
  
• Spatial understanding (GenieSim-Spatial) – Evaluates reasoning over geometric and semantic spatial relationships
  
• Manipulation abilities (GenieSim-Manip) – Assesses execution of atomic abilities and long-horizon activity composition
  
• Robustness (GenieSim-Strong) – Exams adaptability below real-world disturbances equivalent to lighting adjustments, sensor noise, and atmosphere variations
  
• Sim2Real (GenieSim-Sim2Rea) – Features a sequence of analysis duties for zero-shot real-robot switch with excessive success charges
  

GenieSim x RLinf: Scaling reinforcement studying in simulation

Genie Sim 3.0 additionally introduces deep integration with the RLinf framework, enabling an entire reinforcement studying (RL) pipeline for embodied AI.

AGIBOT stated this enhances vision-language-action (VLA) fashions, utilizing low-cost RL post-training to bridge the final mile from “generalized understanding” to “exact micromanipulation.” It listed the next options:

• Decoupled physics and rendering engines – Helps high-frequency (1,000Hz) physics simulation alongside high-fidelity visible commentary
• Massively parallel simulation – Considerably will increase information throughput and accelerating mannequin convergence
• Closed-loop coaching and analysis – RL brokers could be skilled and evaluated instantly inside Genie Sim duties, with built-in reward alerts
• Standardized Health club interfaces – Ensures compatibility with RLinf and broader ecosystem instruments

“This integration allows a seamless pipeline from large-scale simulation coaching to analysis – bridging the hole between normal understanding and exact management,” stated the corporate.

Genie Sim 3.0 integrates with the RLinf framework for a reinforcement studying pipeline. Supply: AGIBOT

AGIBOT builts unified infrastructure for embodied AI

By combining large-scale simulation information, giant language mannequin (LLM)-driven atmosphere era, and standardized analysis, AGIBOT asserted that Genie Sim 3.0 brings collectively the complete growth stack:

Surroundings → Information → Coaching → Analysis

This will considerably cut back the engineering overhead historically required for robotics growth, enabling sooner iteration and broader experimentation, claimed the corporate.

“Because the boundary between simulation and actuality continues to slender—and as atmosphere era scales from hours to minutes—Genie Sim 3.0 offers a crucial basis for the large-scale deployment of embodied AI,” it said.

Open, shared infrastructure like Genie Sim may play a key function in accelerating the evolution of the worldwide robotics ecosystem, stated AGIBOT.

The put up AGIBOT introduces Genie Sim 3.0 simulation platform for embodied AI appeared first on The Robotic Report.