Vision-Language-Action (VLA) models provide a unified framework for grounding high-level semantic information into low-level robot actions, enabling scalable robotic manipulation across diverse tasks. However, existing VLA models lack explicit mechanisms to disentangle the states and intents of the two arms, leading to unintended cross-arm interference that degrades task execution success. To address this issue, we propose a symmetric Dual-Arm Expert (DAE) architecture built upon a shared Vision-Language Model (VLM) backbone with decoupled, arm-specific expert towers. Expert selection is carried out through a two-stage dual-arm intent routing scheme, in which experts are routed either by explicit language instructions in the first stage or by implicit visual semantics in the second stage. Moreover, we introduce a lightweight task progress prediction module that leverages cross-attention between the pre-chunk temporal features and semantic representations of proprioceptive and visual observations to accurately estimate frame-wise task completion progress. This module facilitates task progress synchronization to support coordinated scheduling for collaborative multi-robot tasks. Experimental results demonstrate the effectiveness of our model in dual-arm intent routing and the disentanglement of cross-arm interference, and further provide preliminary evidence of emergent skill generalization from single- to dual-arm tasks (as well as the reverse), together with cross-arm motion-domain skill transfer.
Figures & tables
Fig. 1: Overview of TAO-DA. SigLIP encodes the multi-view observations, and Eagle-2.5-VL fuses visual and language tokens into the context used by DAIR. After routing, inactive end-effector-view tokens are masked at the VLM input when constructing the action-conditioning context. DAE then generates actions through the selected arm-specific pathways, trained with flow matching.
Fig. 2: Architectural variants of DAE. Single-Tower uses a unified state and action space, Dual-Mask introduces weak decoupling through selective token masking in a shared DiT, and Dual-Tower provides strong decoupling through two parameter-independent DiT branches. All variants support single-arm and dual-arm execution.
Category
Task
Mode
Traj.
Duration (h)
Single-arm
Steamer Delivery
Single
1,200
3.0
Bottle Grasping
Single
880
3.2
Fruit Grasping
Single
4,000
14.4
Tea Pouring
Single
1,600
6.0
Dual-arm
Collaborative Tea
Dual
500
2.0
Desktop Organization
Dual
900
3.3
TABLE I: Real-world teleoperation dataset.
Task
GR00T
π0
π0.5
smolVLA
GO-1
TAO-DA
Bottle
85
92.5
95
45
30
95
Steamer-L
90
–
90
–
57.5
94
Steamer-R
97.5
–
75
–
45
95
Green Tea-L
66.7
–
75
–
12.5
75
Black Tea-R
70.8
–
67.5
–
29.2
83
Average
82.0
–
80.5
–
34.8
88.4
TABLE II: Single-arm task success rates (%). The average includes only models evaluated on all five tasks.
Fig. 3: Active-arm trajectory predictions when the inactive arm is perturbed.
Active arm
Model
Clean MSE
Disturbed MSE
Ratio
Right
Single-Tower
2.69×10−2
5.23×10−2
1.94
Dual-Mask
2.54×10−2
3.73×10−2
1.46
Dual-Tower
2.83×10−2
2.80×10−2
0.99
Left
Single-Tower
5.53×10−2
8.85×10−2
1.60
Dual-Mask
5.88×10−2
6.67×10−2
1.13
Dual-Tower
8.54×10−2
8.54×10−2
1.00
TABLE IV: Offline robustness to inactive-arm disturbances.
Condition
Active arm
Single- Tower
Dual- Mask
Dual- Tower
Nominal
Left
80
90
95
Right
70
95
100
Random initial state
Left
40
85
95
Right
60
80
95
Persistent disturbance
Left
25
65
100
Right
20
65
95
TABLE V: Real-robot success rates (%) under the nominal condition, random initial states, and persistent inactive-arm disturbances (20 trials per condition).
Fig. 4: Qualitative comparison on Desktop Organization from identical initial conditions. The objects are located in the robot’s right-arm workspace. Without arm-intent routing, the inactive left arm attempts an empty grasp; with DAIR, it remains stationary while the right arm executes the task.
Fig. 5: Row-normalized confusion matrix for left-only, right-only, and dual-arm intent classification. Each row reports percentages over the corresponding ground-truth class.
Fig. 6: Per-class routing accuracy under instruction-diversity and task-progress ablations. Multi samples semantically equivalent instructions during training, whereas Fixed uses one instruction per task; w/o Progress removes the task-progress module.
Figure 11
Fig. 9: Qualitative single-to-dual-arm composition from separate arm-specific demonstrations. The two learned skills are executed in parallel without target-task bimanual demonstrations or additional fine-tuning.
Fig. 10: Representative tasks in the autonomous restaurant deployment.
Vision-language-action (VLA) models show strong capabilities in single and dual-arm robotic manipulation. Prior works show coordinated bimanual behaviors can emerge from end-to-end learning, leveraging large vision-language backbones with continuous action prediction. However, as bimanual tasks become tightly coupled and execution constraints become critical, implicit coordination alone is insufficient to ensure reliable, interpretable, and stable behavior. In this work, we propose Co-VLA, a coordination-aware bimanual manipulation framework introducing explicit structural priors into VLA models. We instantiate our method on a state-of-the-art vision-language backbone by replacing its monolithic action head with a Structured Action Expert (SAE) designed for bimanual coordination. Specifically, we introduce explicit structure at the action generation level with a modular coordination-aware loss that shapes shared and residual latents according to task-specific structures. The shared latent encodes task-level coordination intent, while residual latents capture execution adjustments for each arm. At deployment, a Latent-Aware Controller (LAC) interprets the learned representations to modulate synchronization strength, execution asymmetry, smoothness, and safety constraints in real time. LAC operates at the joint-command level and remains compatible with standard control pipelines without requiring force or impedance control. Experiments across simulation and real-world benchmarks show Co-VLA significantly outperforms monolithic baselines, achieving a 27% success rate gain in tight-coordination tasks, more than doubling performance in OOD real-world scenarios (from 13% to 27%), and reducing task completion time by up to 25%.
Yandong Wang, Jiaqian Yu, Xiongfeng Peng +8
Donghua University, Shanghai, China · Samsung R&D Institute China-Beijing (SRCB), China · Samsung AI Center, DS Division, South Korea
In dynamic environments such as warehouses, hospitals, and homes, robots must seamlessly transition between gross motion and precise manipulations to complete complex tasks. However, current Vision-Language-Action (VLA) frameworks, largely adapted from pre-trained Vision-Language Models (VLMs), often struggle to reconcile general task adaptability with the specialized precision required for intricate manipulation. To address this challenge, we propose DAM-VLA, a dynamic action model-based VLA framework. DAM-VLA integrates VLM reasoning with diffusion-based action models specialized for arm and gripper control. Specifically, it introduces (i) an action routing mechanism, using task-specific visual and linguistic cues to select appropriate action models (e.g., arm movement or gripper manipulation), (ii) a dynamic action model that fuses high-level VLM cognition with low-level visual features to predict actions, and (iii) a dual-scale action weighting mechanism that enables dynamic coordination between the arm-movement and gripper-manipulation models. Across extensive evaluations, DAM-VLA achieves superior success rates compared to state-of-the-art VLA methods in simulated (SIMPLER, FurnitureBench) and real-world settings, showing robust generalization from standard pick-and-place to demanding long-horizon and contact-rich tasks.
Xiongfeng Peng, Jiaqian Yu, Dingzhe Li +8
Advanced Research Lab, Samsung R&D Institute China-Beijing (SRCB), China · Samsung AI Center, DS Division, South Korea · Hanyang University ERICA, South Korea
Multi-arm collaboration is becoming a core capability in embodied manipulation. Recent vision-language-action (VLA) models integrate perception, language, and control, but most represent language as a single global instruction and do not provide an explicit mechanism for assigning and composing arm-specific behaviors. This design limits transfer to collaboration patterns that differ from those observed during training. We present MA-VLA, a unified framework for multi-arm collaboration via atomic action assignment. MA-VLA decomposes cooperative behavior into mid-level atomic prompts and allocates them to individual arms, enabling explicit subgoal specification and compositional reuse across tasks. To reduce reliance on fixed execution roles, we introduce Arm Shuffle, a training-time permutation of the observation, state, and assigned atomic prompts for each arm. This permutation enforces role-agnostic instruction following and supports recomposition into unseen coordination patterns, which we term multi-arm compositional generalization. We also construct a benchmark in which test-time collaboration patterns are absent in training set. Across simulation and real-world evaluations, prior state-of-the-art VLAs largely fail under these unseen collaborations, while MA-VLA consistently succeeds. These results indicate that structured, per-arm atomic action assignment offers a practical route to scalable generalization in multi-arm embodied systems. Code, models, and data are available at https://github.com/zhangzaibin/future-robots
Zaibin Zhang, Junlan Xiao, Zhongbo Zhang +11
Dalian University of Technology · Shanghai Jiao Tong University · The Chinese University of Hong Kong, Shenzhen +4