EEmbodied AI Hub
IndustryDeploymentFeaturedEditor’s pick

Video Generation Is Not Planning: Georgia Tech's Danfei Xu Proposes Factor Graphs to Bridge the Action Gap

Danfei Xu argues that high-fidelity video generation does not equal physical planning. He proposes compositional world models using factor graphs and temporal attention to enable test-time skill composition, improving success on out-of-distribution tasks. The approach reveals a video-action gap where action models lag behind visual generalization.

compositional world modelsvideo-action gapfactor graphsfeed:leiphoneGeorgia TechdeploymentDanfei XuRSS 2026

Source: 雷峰网 · July 28, 2026

Share this article so more people can see it

At RSS 2026, Georgia Tech assistant professor and NVIDIA researcher Danfei Xu delivered a sharp critique of the prevailing belief that high-fidelity video generation equates to robot planning. In his talk "Compositional World Models," Xu argued that the ability to generate a visually perfect future video does not mean the robot can physically execute the corresponding actions—a gap he calls the "video-action gap."

Xu's core insight is that generation and planning are fundamentally different. Generation samples high-probability outcomes from a data distribution, while planning must construct valid solutions for novel, long-horizon tasks under physical constraints—solutions that may not exist in the training data. For example, a robot trained on separate skills like "carry heavy object," "open fridge," and "place on shelf" may fail when asked to combine them in a new sequence, such as trying to open the fridge while holding the object.

To bridge this gap, Xu proposes compositional world models inspired by Marvin Minsky's modularity. The key idea is to decompose complex tasks into reusable skill modules that can be dynamically composed at test time. Instead of linear chaining, which suffers from inefficient information propagation, Xu uses factor graphs to define spatial and temporal constraints across skills. This allows, for instance, coordinating two robot arms without dedicated dual-arm training data—each arm is trained individually, and the factor graph enforces constraints like relative pose and synchronization at test time.

Xu's team also introduced a metric called "temporal ratio" to diagnose the video-action gap. They found that during reaching phases, the model relies heavily on future trajectory predictions (high temporal ratio), but during grasping, attention shifts to the current frame for tactile and visual feedback. This insight enables policy guidance: dynamically adjusting the model's attention between future prediction and current observation to improve success on out-of-distribution tasks.

During the Q&A, Xu addressed the role of failure data. He noted that most video models are trained on successful human demonstrations, lacking failure examples. Once planning algorithms can leverage failure predictions, such data will become valuable for safety boundaries. He also argued for keeping world models and policies separate: world models need to ingest massive, diverse data (including failures) to understand physics, while policies benefit from high-quality success data for fine-tuning. Merging them, he said, would increase pre-training convergence difficulty.

For founders and operators, Xu's work highlights a critical bottleneck in embodied AI: the action side is data-starved compared to vision. Building systems that can compose skills at test time—rather than relying on end-to-end generation—may be a more practical path to robust deployment. The factor graph approach offers a way to reuse existing skill libraries without retraining for every new task combination.

Source: 雷峰网.

Related resources on this hub

Jump to projects, models, or datasets mentioned or closely related.

Discussion

Tell us what you think — comments make stories more useful for builders and founders.

Tell us what you think!

Video Generation Is Not Planning: Georgia Tech's Danfei Xu Proposes Factor Graphs to Bridge the Action Gap

Have an account? Log in to use your display name and avatar.

Email is optional and never shown on the page.

More insights

IndustryDeployment雷峰网

Xingchi Power and PKU Healthcare: A Pragmatic Path for Embodied AI in Hospitals

Xingchi Power and PKU Healthcare are collaborating on a three-step pipeline—simulation training, scenario validation, and hospital deployment—to bring embodied AI into medical settings safely. The partnership leverages Xingchi's world model engine and data headset to create high-fidelity virtual hospitals, addressing the lack of training environments and expert data. This pragmatic approach prioritizes safety over speed, focusing on tasks like medication sorting and delivery.

Read
IndustryDeployment雷峰网

RSS 2026 Debate: Data Source, Control Modality Split Robotics Community

At RSS 2026, five experts debated whether world models and policy models should be unified or separate, whether control should rely on pixels or actions, and whether training data should come from the internet or embodied robots. The 'separate' camp won the first round, 'pixel' vs 'action' ended in a draw, and 'internet data' won the third round 61% to 39%. Key takeaways: modular design is preferred for debugging, video alone is insufficient for contact-rich tasks, and hybrid data strategies are

Read
IndustryDeploymentFeatured雷峰网

RSS 2026 Test of Time Award: Morphology as Computation Resets Embodied AI Priorities

The RSS 2026 Test of Time Award for the RBO Hand validates that passive mechanical design can replace compute-heavy sensor suites, signaling a strategic pivot for embodied AI startups. The award highlights environmental constraint exploitation and extreme cost advantages, but high-speed dynamic deployment remains unproven. Founders should reconsider hardware-first approaches over pure algorithmic scaling.

Read