Reka releases inverse dynamics model
RIDM eliminates the need for expensive low-level action labeling in video training data, significantly reducing the barrier to building more realistic AI systems. Developers and robotics researchers can now leverage this open-source model to create more capable AI agents that bridge the simulation-to-reality gap, with immediate access to the weights on Hugging Face.
Reka has released the Inverse Dynamics Model (RIDM), a compact model trained entirely on video games.
RIDM extracts low-level camera and motor commands directly from raw video without requiring labeled data.The model generalizes to real-world footage and addresses the limitations of standard simulated environments that lack real-world physics and noise.
The weights are open-sourced under Apache 2.0 license on Hugging Face.
Standard simulated environments lack real-world physics and noise. Interactive World Models get closer, but training them needs video labelled with low-level actions, and real footage rarely has those labels.
Today, we’re releasing the Reka Inverse Dynamics Model (RIDM), a compact model that extracts low-level camera/motor commands directly from raw video. 🕹️Trained entirely on video games 🌐Generalizes to real-world footage 🔐 Weights open-sourced under Apache 2.0
Read the research note and grab the weights on 🤗Hugging Face: https://t.co/T4bvGkkoEm
#MachineLearning #Robotics #WorldModels #AIResearch #OpenSource
This report is categorized under AITOP's "Models" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include RIDM, Reka, video generation, open-source models.
Source attribution: curated from Reka (@RekaAILabs).