Pinned post
Today, RLWRLD unveils RLDX-1 β our proprietary Robotics Foundation Model.
Across all 8 public benchmarks, RLDX-1 outperforms leading SOTA models including #NVIDIA #GR00T and Physical Intelligence #Ο0 β delivering state-of-the-art performance among open robotics foundation models.
π― A 'Dexterity-First' Philosophy
The industry assumes dexterity will follow once intelligence is solved. We see it the other way around.
Dexterity isn't downstream of intelligence β it's the path intelligence must take to act in the physical world. Real industrial work with five-finger robotic hands depends on signals vision alone can't capture: force (torque), tactile feedback, and the precise moment of contact.
π§ MSAT β Multi-Stream Action Transformer
Where conventional VLAs collapse every input into a single transformer stream, MSAT gives each modality β vision, language, action, touch, memory β its own dedicated stream, then unifies them through joint attention. Force, tactile signals, and long-term memory are handled by purpose-built Physics and Memory modules.
The result: one model that can see, feel, remember, and adapt.
π Performance Highlights
RoboCasa Kitchen β 70.6: the first VLA model to cross the 70-point threshold
GR-1 Tabletop β 58.7: +10.7 percentage points over NVIDIA GR00T N1.6
LIBERO-Plus β 86.7%: top score across 7 robustness variables
Pot-to-Cup Pouring on WIRobotics ALLEX β 70.8%: nearly 2Γ the comparison models, which remained in the high-30% range.
We're also releasing DexBench β our industry-grounded benchmark for dexterous manipulation, defined across five domains: Grasp Diversity, Spatial Precision, Temporal Precision, Contact Precision, and Context Awareness.
π Open Release
Three checkpoints (8.1B parameters each), live now on GitHub and Hugging Face:
RLDX-1-PT β pre-training
RLDX-1-MT-ALLEX β mid-training for ALLEX
RLDX-1-MT-DROID β mid-training for DROID
βοΈ Built on NVIDIA's Cloud-to-Edge Stack
Training and simulation on Isaac GR00T, Isaac Lab, Isaac Sim, and cuRobo. Compute on NVIDIA H100 and A100 GPUs. Edge inference on Jetson AGX Thor with TensorRT. Our collaborations with NVIDIA, AWS, and Microsoft continue across both research and deployment.
π What's Next: The 4D+ World Model
Video-based world models will never surface what isn't in the pixels β contact torque, tactile signals, robot state. Our 4D+ World Model integrates these directly with vision, language, and action across the temporal dimension, predicting and generating the full physical world. RLDX-1 is the first milestone on that roadmap.
π Join us at Dexterity Night in San Francisco on May 13 β followed by launch events in Japan and Korea.
π Explore RLDX-1 on GitHub and Hugging Face.
rlwrld.ai/ko/rldx-1
#RLWRL #RLDX1 #PhysicalAI #RoboticsFoundationModel #VLA #Humanoid #Dexterity #FoundationModel #Robotics #AI
Across all 8 public benchmarks, RLDX-1 outperforms leading SOTA models including #NVIDIA #GR00T and Physical Intelligence #Ο0 β delivering state-of-the-art performance among open robotics foundation models.
π― A 'Dexterity-First' Philosophy
The industry assumes dexterity will follow once intelligence is solved. We see it the other way around.
Dexterity isn't downstream of intelligence β it's the path intelligence must take to act in the physical world. Real industrial work with five-finger robotic hands depends on signals vision alone can't capture: force (torque), tactile feedback, and the precise moment of contact.
π§ MSAT β Multi-Stream Action Transformer
Where conventional VLAs collapse every input into a single transformer stream, MSAT gives each modality β vision, language, action, touch, memory β its own dedicated stream, then unifies them through joint attention. Force, tactile signals, and long-term memory are handled by purpose-built Physics and Memory modules.
The result: one model that can see, feel, remember, and adapt.
π Performance Highlights
RoboCasa Kitchen β 70.6: the first VLA model to cross the 70-point threshold
GR-1 Tabletop β 58.7: +10.7 percentage points over NVIDIA GR00T N1.6
LIBERO-Plus β 86.7%: top score across 7 robustness variables
Pot-to-Cup Pouring on WIRobotics ALLEX β 70.8%: nearly 2Γ the comparison models, which remained in the high-30% range.
We're also releasing DexBench β our industry-grounded benchmark for dexterous manipulation, defined across five domains: Grasp Diversity, Spatial Precision, Temporal Precision, Contact Precision, and Context Awareness.
π Open Release
Three checkpoints (8.1B parameters each), live now on GitHub and Hugging Face:
RLDX-1-PT β pre-training
RLDX-1-MT-ALLEX β mid-training for ALLEX
RLDX-1-MT-DROID β mid-training for DROID
βοΈ Built on NVIDIA's Cloud-to-Edge Stack
Training and simulation on Isaac GR00T, Isaac Lab, Isaac Sim, and cuRobo. Compute on NVIDIA H100 and A100 GPUs. Edge inference on Jetson AGX Thor with TensorRT. Our collaborations with NVIDIA, AWS, and Microsoft continue across both research and deployment.
π What's Next: The 4D+ World Model
Video-based world models will never surface what isn't in the pixels β contact torque, tactile signals, robot state. Our 4D+ World Model integrates these directly with vision, language, and action across the temporal dimension, predicting and generating the full physical world. RLDX-1 is the first milestone on that roadmap.
π Join us at Dexterity Night in San Francisco on May 13 β followed by launch events in Japan and Korea.
π Explore RLDX-1 on GitHub and Hugging Face.
rlwrld.ai/ko/rldx-1
#RLWRL #RLDX1 #PhysicalAI #RoboticsFoundationModel #VLA #Humanoid #Dexterity #FoundationModel #Robotics #AI
www.rlwrld.ai RLDX-1 νμ΄λ°μ΄μ
λͺ¨λΈ | RLWRLD λ€μ― μκ°λ½ λ‘λ΄ μ‘°μμ μν΄ κ°λ°λ RLWRLDμ λ±μ€ν°λ¦¬ν° μ€μ¬ νμ΄λ°μ΄μ
λͺ¨λΈ RLDX-1μ μκ°ν©λλ€. 
