Dynin Robotics uses shared model to improve robot action and goal prediction
Dynin-Robotics: Omnimodal Unified Diffusion Vision-Language-Action Model
RoboticsArtificial IntelligenceMachine Learning
Summary
Robots need to understand both what to do and how their actions change the world to succeed at tasks described by language. The authors developed Dynin-Robotics, a single model that learns to predict robot actions, future scenes, goals, and instructions all together. It does this by turning everything into tokens and training on a large variety of robot data. This unified approach helps robots perform better in new tasks and speeds up decision-making. The model was tested successfully on several robot benchmarks and a real robot arm.
What this means in practice
- •For robot control engineers: Improve robot task performance by using a single model that predicts actions, goals, and future states together for better decision making from language inputs.
- •For manufacturing automation teams: Deploy robot arms that adapt more reliably to varied instructions and conditions by leveraging unified trajectory and goal modeling.
Authors
Hoeun Lee, Jaeik Kim, Jusang Oh, Jinhyeok Kim, Geon Choi, Hyeonggeun Kim, Jaeyoung Do
Abstract
Visual goal and dynamics prediction can provide language-conditioned robot policies with both a target outcome and a representation of action-dependent scene changes. We bring these predictions into action generation and selection through a shared trajectory model. Dynin-Robotics implements this formulation on Dynin-Omni, an omnimodal masked-diffusion backbone, representing language, visual observations, goals, and actions as discrete tokens. By varying conditioning and target spans, the same model learns action prediction, action-conditioned next-observation prediction, terminal goal-state prediction, and trajectory-to-instruction reconstruction. These interfaces support test-time scaling through goal prediction, action-candidate evaluation, and joint refinement of action and future-state predictions. We continually pretrain the model on approximately 1.33 million trajectories from 48 Open X-Embodiment datasets and adapt it separately to downstream domains. On two VLABench tasks, robot pretraining improves adaptation within a fixed Stage-2 step budget, and the full objective mixture improves shifted-instruction success over Policy-only post-training under the same coupled decoder. Combining goal guidance with joint action-next-state denoising further improves shifted-instruction success over action-only decoding; the benefit depends on how the predictions are composed. Dynin-Robotics achieves competitive performance on LIBERO and zero-shot LIBERO-Plus, together with a 78.4% average success rate across four manipulation conditions on a Franka Research 3 robot. An optimized block-parallel implementation accelerates model-side action decoding by up to 29.2x relative to the base implementation under the reported profiling setup. These results support shared trajectory modeling as a common interface for learning complementary robot objectives and composing their predictions during control.