$ω$-EVA: Envision, Verify, and Act with Latent Interactive World Models
2026-06-08 • Robotics
Robotics
AI summaryⓘ
The authors present ω-EVA, a system that helps a robot plan actions by first imagining what might happen if it acts a certain way, then checking that imagined outcome before deciding what to do. Unlike usual methods that choose actions without looking ahead, ω-EVA uses a 'world model' to predict future states inside the system without creating full videos. The system improves decision-making by combining current observations, imagined futures, and proposed actions to pick better moves. Tests show that this approach works well across various robot tasks without needing extra robot-specific training.
embodied policiesworld modellatent dynamicsflow policyvisual representationsaction-conditioned predictionEnvision–Verify–Act looprobot simulationlatent feature spacerobot action planning
Authors
Zhenguo Sun, Yu Sun, Hande Huang, Alois Knoll
Abstract
Embodied policies typically map current observations directly to actions, leaving candidate-action consequences implicit. World models provide predictive supervision, representations, or external simulation, but rarely let a policy inspect the imagined consequence of its own proposal before acting. We introduce $ω$-EVA, a latent interactive world model that realizes an Envision--Verify--Act loop for embodied action generation. Its three-stage framework learns action-conditioned latent dynamics, trains a language-conditioned flow policy on dynamics-aware visual representations, and feeds the policy's proposal back through the world model. A tri-branch refiner jointly reasons over the current state, proposal-conditioned future, and proposed action to produce the final action chunk. Because consequence reasoning remains in latent feature space, $ω$-EVA avoids generating future videos at inference. Evaluations across diverse single-arm, bimanual, long-horizon, and perturbed simulation settings show that the complete interaction pipeline consistently improves the proposal policy, while latent diagnostics indicate meaningful action-conditioned future structure. With approximately 1.2B parameters and no additional robot-data pretraining, $ω$-EVA demonstrates a compact and competitive performance--scale--data trade-off, making the world model an active action-feedback module rather than a passive predictor.