cs.ROSep 13, 2026
SaveVLBiMan++: Expanding the Generalization Boundary of Vision-Language Anchored One-Shot Bimanual Manipulation
Organizations: Guangdong Provincial Key Laboratory of Visual Media and Multidimensional Intelligence, College of Computer Science and Software Engineering, Shenzhen University
Abstract
Generalizable bimanual robotic manipulation requires a reusable task prior that can persist across increasingly diverse tasks, objects, scenes, embodiments, and execution conditions, thus avoiding the prohibitive cost of large-scale teleoperated demonstrations and policy retraining. In this work, we present VLBiMan++, an extended framework that expands the generalization boundary of vision-language anchored one-shot bimanual manipulation. Starting from a single human demonstration, VLBiMan++ performs task-aware decomposition to identify reusable and adaptable skill components, and employs vision-language grounded geometric adaptation to transfer these skills to novel configurations without retraining. Building on this foundation, we systematically extend generalization along five dimensions: task generalization through diverse and long-horizon skill compositions; object generalization across unseen categories, varying geometries, and more complex articulated or deformable objects; scene generalization under clutter, occlusion, and dynamic interference; embodiment generalization across heterogeneous dual-arm robotic platforms; and deployment generalization through prolonged closed-loop execution under repeated external perturbations. To support this broader scope, we further introduce object-state-aware adaptation and lightweight trajectory optimization mechanisms that accommodate changes beyond simple rigid 6-DoF pose variations while preserving reliable bimanual coordination. Extensive real-world experiments demonstrate that VLBiMan++ maintains strong task success and adaptation capability across these increasingly challenging settings. Overall, VLBiMan++ advances one-shot bimanual manipulation from demonstrating isolated transferability toward a more systematic and scalable framework for generalization across tasks, objects, scenes, embodiments, and long-term deployment conditions.
Figures & tables
Fig. 1: Left : Taking pouring water as an example, we sketch the entire process of VLBiMan++ based on the one-shot demonstration. Right : Diagram illustration of the proposed Generalization Pyramid : Task Object Scene Embodiment Deployment . VLBiMan++ aims to investigate how far a single human demonstration can serve as a reusable manipulation prior across increasingly diverse tasks, objects, scenes, embodiments, and execution conditions.
Fig. 2: Framework of the expanded V ision- L anguage Anchored Bi manual Man ipulation ( VLBiMan++ ). Taking the pouring water as an example, the paradigm consists of three stages ( e.g. , decomposition, adaptation, and composition) based on a given demonstration. VLBiMan++ can achieve generalization of unseen spatial placements, category-level new instances, non-rigid object state variations, and heterogeneous novel embodiments under the same task.
Fig. 3: Illustrations of representative points for target objects in four tasks including pouring , reorient+unscrew , tool-use:spoon , and zip penbag . The first three tasks involve only rigid objects, while the last task involves an articulated object. These anchor points will be used to calculate the change in object position and orientation (not always required).
Algorithm 1 Orientation Estimation for Rigid Objects
Fig. 4: Illustrations of extracted various anchor points for deformable objects in two tasks including coil cable (top) and fold towel (bottom). These can serve as a supplement to Alg. 2 . We present the computational results for four test cases of each task. In addition, we also present intermediate results for handling linear or rectangular deformable objects.
| primary bimanual tasks/skills | long-horizon multi-stage tasks | new rearrangement tasks | new non-rigid targets | ||||||||||||||||
| Task Types & Names | plugpen | inserting | unscrew | pouring | pressing | handover | reorient +unscrew | unscrew +pouring | tool-use scoop | tool-use funnel | flatting | reorient | flipping | upright | place bottle_mug | place fork_spoon | zip penbag | coil cable | fold towel |
| Dual-Arm Platform(s) | P1 P2 | P1 P2 | P1 P2 | P1 P2 | P1 | P1 P2 | P1 | P1 P2 | P1 | P1 | P2 | P2 | P2 | P2 | P2 | P2 | P2 | P2 | P2 |
| Object Number | 2 | 2 | 1 | 2 | 2 | 1 | 1 | 2 | 3 | 3 | 1 | 1 | 1 | 1 | 2 | 2 | 2 | 1 | 1 |
| Object Names /States | pen body +cap | pen +cup | upward bottle | upward bottle +mug | cup+ pump bottle | spoon or shovel | lying bottle | upward bottle +mug | spoon+ bowl_1+ bowl_2 | funnel +bottle +mug | upside down bottle | lying bottle | upside down mug | lying mug | bottle +mug | fork+ spoon | penbag +pen | linear cable | rectan -gular towel |
| Object Properties | Art. | Rig. | Rig. | Rig. | Rig. | Rig. | Rig. | Rig. | Rig. | Rig. | Rig. | Rig. | Rig. | Rig. | Rig. | Rig. | Art. | Def. | Def. |
| Orientation Estimation? | ✓✓ | ✓✗ | ✗ | ✗✗ | ✗✗ | ✓ | ✓ | ✗✗ | ✓✗✗ | ✓✗✗ | ✗ | ✓ | ✗ | ✓ | ✓✓ | ✓✓ | ✓✓ | ✗ | ✗ |
TABLE I: Detailed statistics on all bimanual manipulation tasks, including 10 previous tasks and 9 newly added tasks. In the Dual-Arm Platform(s) , P1 and P2 represent the task has been instantiated on a contralateral robot and a humanoid robot, respectively. In the Object Properties , Rig., Art., and Def. represent rigid, articulated, and deformable objects, respectively. Among them, deformable objects require Alg. 2 to extract anchor points. And the Orientation Estimation is based on Alg. 1 .
| six primary bimanual tasks/skills | four long-horizon multi-stage tasks | |||||||||||||
| Test Setting | Dyna -mic Inter- ference | Manipulation Method | plugpen | inserting | unscrew | pouring | pressing | handover | Average Success Rate | reorient +unscrew | unscrew +pouring | tool-use scoop | tool-use funnel | Average Success Rate |
| ID | No | Mechanisms | 11/25 | 09/25 | 05/25 | 05/25 | 07/25 | 03/25 | 26.7% | 05/25 | 04/25 | 02/25 | 01/25 | 12.0% |
| MAGIC | 16/25 | 15/25 | 10/25 | 10/25 | 09/25 | 07/25 | 44.7% | 09/25 | 08/25 | 04/25 | 03/25 | 24.0% | ||
| Robot-ABC | 14/25 | 10/25 | 09/25 | 07/25 | 08/25 | 06/25 | 36.0% | 06/25 | 06/25 | 03/25 | 03/25 | 18.0% | ||
| ReKep | 14/25 | 11/25 | 10/25 | 12/25 | 10/25 | 08/25 | 43.3% | 07/25 | 08/25 | 05/25 | 03/25 | 23.0% | ||
| ReKep+ | 19/25 | 18/25 | 13/25 | 17/25 | 17/25 | 11/25 | 63.3% | 11/25 | 10/25 | 07/25 | 06/25 | 34.0% | ||
TABLE II: Quantitative comparison results of success rates on six primary bimanual tasks/skills and four long-horizon multi-stage tasks under the in-distribution (ID) setting (new placements + same objects ) or the out-of-distribution (OOD) setting (new placements + novel instances ) using the contralateral dual-arm robot ( P1 ).
Fig. 7: Visualizations of ten tasks executed on the contralateral dual-arm robots platform P1 . They are designed to validate different aspects, including (a) six dual-arm primary skills, (b) combination of basic skills for two long-horizon tasks, and (c) exploration of two multi-stage tool-use tasks.
| six new rearrangement tasks | three new non-rigid targets | ||||||||||||
| Test Setting | Dyna -mic Inter- ference | Manipulation Method | flatting | reorient | flipping | upright | place bottle_mug | place fork_spoon | Average Success Rate | zip penbag | coil cable | fold towel | Average Success Rate |
| ID | No | Mechanisms | 04/25 | 04/25 | 02/25 | 01/25 | 01/25 | 01/25 | 8.7% | 00/25 | 00/25 | 02/25 | 2.7% |
| MAGIC | 06/25 | 05/25 | 03/25 | 02/25 | 01/25 | 02/25 | 12.7% | 00/25 | 00/25 | 03/25 | 4.0% | ||
| Robot-ABC | 07/25 | 06/25 | 06/25 | 06/25 | 05/25 | 05/25 | 23.3% | 01/25 | 00/25 | 05/25 | 8.0% | ||
| ReKep+ | 13/25 | 12/25 | 10/25 | 11/25 | 08/25 | 11/25 | 43.3% | 04/25 | 00/25 | 08/25 | 16.0% | ||
| VLBiMan++ | 24/25 | 22/25 | 18/25 | 20/25 | 15/25 | 23/25 | 81.3% | 12/25 | 10/25 | 20/25 | 56.0% | ||
TABLE III: Quantitative comparison results of success rates on six new rearrangement tasks and three new non-rigid targets under the in-distribution (ID) setting (new placements + same objects ) or the out-of-distribution (OOD) setting (new placements + novel instances ) using the humanoid dual-arm robot ( P2 ). The ’—’ means there are no extra new object instances.
Fig. 8: Visualizations of newly added nine tasks executed on the humanoid-style dual-arm platform P2 . They are defined to further confirm the universality of VLBiMan++, including (a) four left-arm or right-arm rearrangement skills, (b) combination of new basic skills for two long-horizon rearrangement tasks, (c) a complex task involving articulated bodies, (d) two dual-arm tasks involving linear or rectangular deformable objects.
Fig. 9: Visualizations of VLBiMan++’s robustness performance in cluttered scenarios on the platform P2 . We selected three tasks for extensive evaluation: ( a ) reorient , ( b ) pouring , and ( c ) inserting . The red arrow indicates the direction of the manually moved or deliberately obscured object (interfering). The cyan arrow and yellow arrow indicate the movement direction of the left and right robotic arms (chasing) respectively.
| Dynamic Interference | ID setting: new placements + same objects | OOD setting: new placements + novel instances | |||||||||||||
| Dual-Arm Platform Type | plugpen | inserting | unscrew | pouring | handover | unscrew +pouring | Average Success Rate | plugpen | inserting | unscrew | pouring | handover | unscrew +pouring | Average Success Rate | |
| No | Contralateral | 19/20 | 18/20 | 16/20 | 17/20 | 15/20 | 12/20 | 80.8% | 15/20 | 17/20 | 14/20 | 14/20 | 14/20 | 09/20 | 69.2% |
| Humanoid | 20/20 | 19/20 | 17/20 | 16/20 | 15/20 | 13/20 | 83.3% | 16/20 | 18/20 | 16/20 | 13/20 | 14/20 | 09/20 | 71.7% | |
| Yes | Contralateral | 16/20 | 13/20 | 15/20 | 15/20 | 12/20 | 08/20 | 65.8% | 14/20 | 11/20 | 12/20 | 11/20 | 10/20 | 07/20 | 54.2% |
| Humanoid | 16/20 | 14/20 | 15/20 | 14/20 | 13/20 | 09/20 | 67.5% | 12/20 | 12/20 | 13/20 | 12/20 | 10/20 | 08/20 | 55.8% | |
TABLE IV: Quantitative results of VLBiMan++’s success rates on six transferred bimanual tasks from the contralateral dual-arm robot ( P1 ) to the humanoid dual-arm robot ( P2 ) under the ID and OOD settings.
Fig. 10: Visualizations of five cross-embodiment transferred primary bimaual tasks executed on the new humanoid-style robot ( P1 P2 ).
| ID setting: new placements + same objects | OOD setting: new placements + novel instances | |||||||||||||
| Manipulation Method | plugpen | inserting | unscrew | pouring | pressing | handover | Average Success Rate | plugpen | inserting | unscrew | pouring | pressing | handover | Average Success Rate |
| Mechanisms | 01/20 | 01/20 | 00/20 | 01/20 | 01/20 | 00/20 | 3.3% | 00/20 | 00/20 | 00/20 | 00/20 | 00/20 | 00/20 | 0.0% |
| MAGIC | 03/20 | 04/20 | 02/20 | 02/20 | 02/20 | 01/20 | 11.7% | 01/20 | 02/20 | 00/20 | 01/20 | 01/20 | 00/20 | 4.2% |
| Robot-ABC | 02/20 | 02/20 | 01/20 | 01/20 | 01/20 | 01/20 | 6.7% | 00/20 | 00/20 | 00/20 | 00/20 | 00/20 | 00/20 | 0.0% |
| ReKep | 05/20 | 03/20 | 02/20 | 01/20 | 02/20 | 02/20 | 12.5% | 03/20 | 01/20 | 01/20 | 01/20 | 01/20 | 00/20 | 5.8% |
| ReKep+ | 08/20 | 06/20 | 04/20 | 03/20 | 04/20 | 05/20 | 25.0% | 05/20 | 02/20 | 03/20 | 02/20 | 03/20 | 01/20 | 13.3% |
TABLE V: Quantitative comparison results of success rates on six primary bimanual skills/tasks using the contralateral dual-arm robot ( P1 ). All tasks faced the challenges of dynamic interference and uneven illumination .
Fig. 11: Video snapshots recorded from a third-person perspective for the super long-duration experiment of reorient on platform P2 . It contains keyframes from 20 trials in which the left arm ( cyan circle) or right arm ( yellow circle) was used to upright a bottle lying down in an arbitrary orientation.
Fig. 12: Examples of interesting findings about the closed-loop mechanism. Top row : this case comes from the pre-grasping phase of pouring , where the left arm approaches and grasps the bottle. Bottom row : this case comes from the pre-grasping phase of inserting , where the right arm approaches and grasps the marker from the top direction.
Fig. 13: Taking the inserting task as an example, we can replace the marker pen held by the left arm with other rectangular objects that were completely different ( e.g. , spoon in a1/a2 , brush in a3/a4 , spatula in a5 , syringe in a6 , and toothbrush in a7/a8 ) as in the top row , or change the cup grasped by the right arm to cups of different shapes ( e.g. , mugs with handles in b1/b2/b3/b4 , and ordinary cups without handles in b5/b6/b7/b8 ) as in the bottom row .
| Inter- polation Density | ID setting | OOD setting | ||||||||
| inserting | unscrew | pouring | reorient | Average Success Rate | inserting | unscrew | pouring | reorient | Average Success Rate | |
| 15/20 | 14/20 | 12/20 | 13/20 | 67.5% | 13/20 | 12/20 | 11/20 | 11/20 | 58.8% | |
| 18/20 | 15/20 | 15/20 | 16/20 | 80.0% | 17/20 | 14/20 | 14/20 | 14/20 | 73.8% | |
| 19/20 | 17/20 | 18/20 | 16/20 | 87.5% | 18/20 | 16/20 | 17/20 | 15/20 | 82.5% | |
| 19/20 | 19/20 | 18/20 | 17/20 | 91.3% | 19/20 | 18/20 | 17/20 | 16/20 | 87.5% | |
| 19/20 | 18/20 | 18/20 | 18/20 | 92.5% | 18/20 | 19/20 | 17/20 | 16/20 | 87.5% | |
TABLE VI: Ablation experiments of the interpolation density . To ensure reliable searching of the optimal , we did not add any additional dynamic interference in each trail.
| VLMs-based grounding | state- aware adaptation | IK- refinement | collision avoidance | closed-loop re-composition | Avg. SR |
| SAM+DINOv2 | Ours | ✓ | ✓ | ✓ | 35.8% |
| Ours | AnyGrasp | ✓ | ✓ | ✓ | 31.7% |
| Ours | Ours | ✗ | ✓ | ✓ | 29.2% |
| Ours | Ours | ✓ | ✗ | ✓ | 34.2% |
| Ours | Ours | ✓ | ✓ | ✗ | 40.8% |
| Ours | Ours | ✓ | ✓ | ✓ | 59.2% |
TABLE VII: Ablation studies of VLBiMan++. All trials were completed on six primary tasks using the platform P1 ( ID + interference ).
Fig. 14: Error breakdown performed on VLBiMan++ failures based on the rollout results in Tab. II (covering OOD and interference part).
More in this topic
Bimanual Robotic Manipulation
Bimanual robotic manipulation coordinates two robot arms to handle objects, tools, and tasks that require collaboration.
See all 178 papers