⚛️量子位•Stalecollected in 34m
Ant Group's LingBot-VA accepted to RSS 2026
💡Learn how Ant Group's LingBot-VA enables robots to reason and act simultaneously, a key milestone for embodied AI.
⚡ 30-Second TL;DR
What Changed
LingBot-VA research accepted by RSS 2026, a top-tier robotics conference.
Why It Matters
This advancement bridges the gap between high-level cognitive reasoning and low-level motor control, potentially accelerating the development of more autonomous and responsive robotic agents.
What To Do Next
Follow the RSS 2026 proceedings to analyze the LingBot-VA architecture for implementing real-time reasoning in your own embodied AI projects.
Who should care:Researchers & Academics
Key Points
- •LingBot-VA research accepted by RSS 2026, a top-tier robotics conference.
- •Enables robots to perform reasoning and physical execution in parallel.
- •Demonstrates significant progress in embodied AI and real-time decision-making.
🧠 Deep Insight
Web-grounded analysis with 5 cited sources.
🔑 Enhanced Key Takeaways
- •LingBot-VA is an autoregressive diffusion framework that unifies video world modeling and policy learning, simultaneously learning future frame prediction and action execution.
- •The model employs a novel approach that first predicts future states (video frames) and then infers the necessary actions to achieve those states, departing from direct observation-to-action mapping.
- •LingBot-VA demonstrates significant improvements in long-horizon tasks, data efficiency (requiring as few as 10-50 demonstrations), and generalization to novel scenes, outperforming state-of-the-art methods on benchmarks like RoboTwin 2.0 and LIBERO.
- •It features a dual-stream Mixture-of-Transformers (MoT) architecture with asynchronous execution and KV Cache, enabling robust long-term memory and efficient parallel inference and execution.
🛠️ Technical Deep Dive
- LingBot-VA is an autoregressive diffusion framework.
- It unifies visual dynamics prediction and action inference within a single interleaved sequence.
- The architecture utilizes a dual-stream Mixture-of-Transformers (MoT).
- It incorporates Asynchronous Execution and KV Cache for high-efficiency execution and long-term memory, allowing the model to remember past actions and progress.
- The framework operates in three stages: (1) Autoregressive video generation predicts future frames conditioned on current observations and language instructions; (2) An inverse dynamics model (IDM) decodes actions from the predicted video; (3) After execution, real observations replace the video KV-cache, grounding the model in actual outcomes for closed-loop control.
- It leverages a 5.3B video diffusion backbone (Wan2.2) as a "world model" to predict future frames.
- Causal attention is integrated to ensure logical rigor, restricting the model to use only past information and prevent peeking into the future.
- Engineering designs include Partial Denoising for action predictions to improve computational efficiency and FDM Grounding to continuously correct the model's imagination with real observation data, preventing hallucination drift.
🔮 Future ImplicationsAI analysis grounded in cited sources
LingBot-VA's paradigm will significantly accelerate the development of more autonomous and adaptable robots for complex real-world applications.
Its ability to perform simultaneous reasoning and physical actions, coupled with strong performance on long-horizon and data-efficient tasks, suggests a clear path towards robots capable of handling intricate, multi-step operations in dynamic environments.
Ant Group's strategy of open-sourcing LingBot-VA and related models will foster broader innovation and adoption of embodied AI technologies.
Making foundational models publicly available can lower barriers to entry for researchers and developers, potentially leading to faster advancements and diverse applications across the robotics community.
⏳ Timeline
2004
Alipay, the predecessor to Ant Group's financial transaction arm, was established.
2014-10
Ant Financial Services Group (now Ant Group) was founded.
2023
Ant Group invested 21.19 billion yuan in technology R&D, primarily in AI, and received approval to release products powered by its 'Bailing' AI large language model.
2024-Q4
Ant Lingbo Technology (Robbyant), Ant Group's robotics unit, was founded in Shanghai.
2025-09-11
Ant Group unveiled its first humanoid robot, R1, developed by Robbyant, at the 2025 Inclusion Conference.
2026-01-29
LingBot-VA, an autoregressive diffusion framework, was introduced in the paper 'Causal World Modeling for Robot Control' and made available on Hugging Face.
2026-04-16
Robbyant open-sourced LingBot-Map, a streaming 3D reconstruction model, as part of its LingBot suite, which includes LingBot-VA.
2026-05-25
Ant Group's LingBot-VA research accepted to the prestigious Robotics: Science and Systems (RSS) 2026 conference.
📎 Sources (5)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 量子位 ↗