DECO: Decoupled Multimodal Diffusion Transformer for Bimanual Dexterous Manipulation with a Plugin Tactile Adapter

Bimanual dexterous manipulation relies on integrating multimodal inputs to perform complex real-world tasks. To address the challenges of effectively combining these modalities, we propose DECO, a decoupled multimodal diffusion transformer that disentangles vision, proprioception, and tactile signals through specialized conditioning pathways, enabling structured and controllable integration of multimodal inputs, with a lightweight adapter for parameter-efficient injection of additional signals. Alongside DECO, we release DECO-50 dataset for bimanual dexterous manipulation with tactile sensing, consisting of 50 hours of data and over 5M frames, collected via teleoperation on real dual-arm robots. We train DECO on DECO-50 and conduct extensive real-world evaluation with over 2,000 robot rollouts. Experimental results show that DECO achieves the best performance across all tasks, with a 72.25\% average success rate and a 21\% improvement over the baseline. Moreover, the tactile adapter brings an additional 10.25\% average success rate across all tasks and a 20\% gain on complex contact-rich tasks while tuning less than 10\% of the model parameters.

Imagine a robot assembling parts with both hands. It can see the parts and sense its own joint positions, but without touch it cannot tell whether two parts have made contact or whether it is pressing hard enough. Humans rely heavily on touch for such tasks, yet most robot AI systems today either ignore touch entirely or mix it with visual data in a way that lets the rich visual signal drown out the sparse touch signal. We propose DECO, a method that gives each sensory input—vision, body awareness, and touch—its own dedicated pathway inside the AI model. This "decoupled" design ensures touch signals are not overwhelmed. We also introduce a lightweight adapter that adds touch sensing to an existing vision-only robot policy by retraining fewer than 10% of its parameters. We collected DECO-50, a 50-hour dataset of bimanual dexterous manipulation with touch sensing. Across over 2,000 real-world trials, DECO outperformed existing methods by 21%, and the touch adapter further boosted success by 20% on contact-intensive tasks. Code and data are publicly released.