NTH

Duo: A Reciprocal World Model for Full-Duplex Interaction

AuthorsHengyuan Cao, Biao Gong, Guoshan Lu, Qiyuan Zhang, Lin Liu, Yiding Tian, Yilei Jiang, Yuxin Ma, Chengzhang Wu, Kun Chen, Yiqian Xie, Ting Liu, Xin Zheng, Xijun Gu, Tianren Ma, Jiancheng Pan, Chi Zhang, Xuzhe Zheng, Siyi Li, Zhenzhong Lan

November 20, 2026 3 min read
Watch on YouTube
The one-line take

Duo aims to create an always-on audiovisual assistant that can perceive, reason, and respond continuously, including during interruptions.

Key results

17
Interaction chunk

Video frames per chunk, processed at 24 frames per second.

650
Pipeline latency upper bound

Upper end of the reported intent, generation, and overhead time in milliseconds.

93.3%
Overall human-evaluation pass rate

168 of 180 criterion checks passed across 30 real-time interaction cases.

73.3%
Content responsiveness

Pass rate for whether replies addressed what users said and showed.

What the paper found

Duo introduces a Reciprocal World Model that treats reasoning and audiovisual behavior as parts of the same interactive world, rather than letting a user steer a passive simulator. Unlike physical-world generators such as NVIDIA Cosmos, Duo v1 continuously listens and watches, then can speak, wait, or interrupt while user input continues. Its Intent Module, based on Qwen3-VL, autoregressively decides whether and what to say; its Realization Module, initialized from MiniMax-H3, generates synchronized speech and video. The modules coordinate on chunks of 17 video frames, using causal attention masks so neither can access future inputs. A key training contribution is dual rollout distillation: the system trains the media generator on its own predicted assistant language and generated audiovisual history, reducing the mismatch between training and live use. Distribution matching distillation produces a four-step generator. The reported intent, generation, and runtime-overhead stages take 330–650 ms, before any additional wait for a sampling window or first-frame decoding. In a human evaluation of 30 real-time cases, 168 of 180 checks passed, for 93.3% overall; content responsiveness was 73.3%, the weakest criterion, while language, speech intelligibility, and visual identity each passed every check. The results show a working approach to full-duplex audiovisual interaction, while highlighting that interpreting user intent from live speech and video remains a major limitation.

Original abstract

In this report, we introduce the Reciprocal World Model (RWM) and present Duo v1, a full-duplex world model that continuously perceives audiovisual inputs while responding to users in real time. The RWM treats intelligence and physical reality as equally fundamental to what constitutes a world. Existing world models largely focus on simulating the latter, leaving reasoning and decision-making as external controls rather than intrinsic parts of the world itself. In RWM, we identify two equally fundamental capabilities: Intent, which provides the cognition underlying world interactions, and Realization, which generates coherent audio-visual behavior. To realize this formulation, Duo v1 introduces three key technical designs: (1) Leveraging autoregressive modeling for continuous interaction reasoning and participation decisions, rather than restricting it to instruction understanding for diffusion models. (2) Constructing chunk-specific conditions and synchronize tokens across Intent and Realization on a unified timeline. Module-specific attention masks enforce causal dependencies both within and across chunks, preventing access to future inputs. (3) Mitigating distribution shift in both modules through dual rollout distillation, training on autoregressively generated Intent tokens and synthesized audiovisual histories from Realization. Both modules continuously perceive real-time audio, video, and text, enabling full-duplex interaction with spontaneous conversations and interruptions. Beyond conversation, users can share real objects, reshape virtual scenes, guide assistants, or play audiovisual games without mouse or keyboard controls. The release of Duo v1 makes our first step, with the next version coming soon. We look forward to broader exploration of this emerging field.

Read the original paper

More in World Models

Browse all 59 papers →
01World Model

PanDWM: Panoramic 4D Driving World Models

Jiajun Sun, Zeyu Zhang, Zeyu Ren, Guangfeng Jiang, Yang Zhao, Hao Tang

PanDWM generates a continuous 360-degree driving view and reconstructs it into a fast 4D scene.

Read analysis
02World Model

WorldRuler: Interactive World Modeling under Configurable Rules

Mingliang Zhai, Chuanhao Li, Jiaming Tan, Zhen Li, Yuwei Wu, Yunde Jia, Zian Meng, Yuanyang Yin, Kaipeng Zhang

WorldRuler aims to make generated interactive worlds obey customizable rules, even after saving and restoring the world.

Read analysis