WasserMan: Benchmark for Underwater Manipulation Policy Learning
Authors: Danil Belov, Artem Erkhov, Sergei Parsegov, Pavel Osinenko
Organizations: Central University, Moscow, Russia · Center for Engineering Systems and Sciences, Moscow, Russia · Sirius University of Science and Technology, Sirius Federal Territory, Russia · V. A. Trapeznikov Institute of Control Sciences of Russian Academy of Sciences, Moscow, Russia
Underwater manipulation couples visual decisions, contact forces and a thruster-controlled floating base. We introduce WasserMan, to our knowledge the first multi-task simulation benchmark for visuomotor learning of floating-base underwater contact manipulation. It provides ten expert-solvable tasks, two vehicle-arm platforms, a bimanual configuration and underwater dynamics. Nine tasks have learned-policy evaluations. We compare ACT, diffusion policies (DP) and behavioral cloning on six tasks, with three training runs and equal sampled-window budgets. Pretrained SmolVLA adds evaluations on three tasks. In the tested settings, removing integral action can prevent completion, while changing action interfaces can reduce learned-policy success despite successful expert replay. Currents produce task-dependent changes in completion and actuation effort. Versioned tasks, demonstrations and per-episode evidence support a protocol separating expert feasibility, learned completion and execution effort.
Figures & tables
Fig. 2: System diagram. Wrist RGB and proprioception feed a policy (shown schematically). Inverse kinematics converts EE targets to joint commands. Native base/joint targets bypass this step. Vehicle feedback control and joint servos drive separate force and torque paths. Allocation and motor response bound thruster forces. Hydrodynamic loads act on the coupled simulation, and simulated contact provides physical measurements for evaluation.
Fig. 3: Six tasks in the learning protocol. The external view (large), base camera (upper right) and gripper camera (lower right) show the same expert state. These CAD-profile views illustrate the tasks. Policy learning uses procedural geometry, wrist RGB and proprioception.
Task
Interface
du
fπ (Hz)
H
PressButton
Actuator
10
30
470
RotateValve
EE pose/jaw
8
30
2240
OpenHatch
Actuator
8
20
1790
CollectShell
Actuator
8
30
7190
PushSlider
Actuator
8
30
1790
PullLever
Actuator
8
30
1790
TABLE I: Decoded command size du , policy target rate fπ , and horizon H in 30 Hz control steps. All families predict 16 targets and execute 8.
Fig. 4: Visual water presets on CabinRecovery (top) and CargoRelease (bottom). Within each row, the camera and physical expert state are identical. The presets combine refraction, attenuation, backscatter and suspended particles. These are optical replays, not learned-policy robustness scores.
Task
ACT
DP
BC
SmolVLA
PressButton
45.6 ± 5.1
60.0 ± 12.0
40.0 ± 5.8
54.4 ± 5.1
RotateValve
84.4 ± 24.1
84.4 ± 6.9
74.4 ± 5.1
—
OpenHatch
81.1 ± 16.8
100.0 ± 0.0
16.7 ± 17.6
—
CollectShell
0.0 ± 0.0
33.3 ± 12.0
0.0 ± 0.0
—
PushSlider
10.0 ± 12.0
42.2 ± 10.2
2.2 ± 1.9
8.9 ± 6.9
PullLever
40.0 ± 38.4
51.1 ± 6.9
24.4 ± 15.0
7.8 ± 5.1
TABLE II: Success (%), mean ± sample SD over three training runs and 30 resets each. ACT, DP and BC cover six tasks. Pretrained SmolVLA covers three of these tasks. Each run uses 320,000 demonstration windows.
Fig. 5: Each point is a separately trained checkpoint evaluated on 30 resets. Within a family, points from left to right are seeds 17, 43 and 101. Horizontal marks show training-run means. Families use the same reset states but independent training draws. SmolVLA is shown for its three-task extension.
kI
Base coefficients
OpenHatch
RotateValve
1
Nominal
30/30/30
25/30/28
0.5
Nominal
17/15/25
21/27/25
0
Nominal
0/0/0
9/9/9
1
Published damping
30/30/30
24/27/27
1
Published added mass
30/30/30
25/28/28
1
Published both
30/30/30
25/28/26
TABLE III: Success counts per 30 research resets, ordered by training seeds 17/43/101. kI multiplies the original integral gain. Published conditions replace only the stated base coefficients. All unlisted settings are unchanged.
Fig. 6: Nominal-model integral intervention. Left: each line is one frozen DP checkpoint’s full-horizon success on 30 research resets. Other panels: change from kI=1 in opposing-grasp occupancy and world-identity attitude RMS over the predeclared 20-s window, fully observed in every pair. Colored markers show mean changes for each training run. Black diamonds and bars show aggregate changes and descriptive 95% intervals from the paired crossed bootstrap. Contact changes have opposite signs across tasks. Interval conventions follow Section IV.
Fig. 7: Complete interface comparison. Each point is one independently trained DP model on 30 research resets. Within each interface, points from left to right are training seeds 17/43/101. Horizontal marks show means. Descriptive crossed intervals use independent training draws between interfaces and paired reset draws. Interface effects differ by task.
Fig. 8: Frozen-policy current intervention. Lines identify training seeds. upper panels report full-horizon completion and lower panels 20 s motor-force RMS. Each task/checkpoint uses the same 30 resets across currents. The three OpenHatch success curves coincide at 100%. Completion and actuator effort respond differently across tasks.
Underwater visuomotor policy learning has focused primarily on single manipulators, while bimanual imitation learning has been studied largely in air. We present ULOHA, an underwater bimanual robot learning platform that combines custom-designed leader--follower hardware with software extensions to LeRobot, integrating teleoperation, multi-view sensing, demonstration collection, policy training, and autonomous deployment. Real-robot experiments demonstrate a range of coordinated underwater bimanual behaviors, including inter-arm transfer, shared-object manipulation, and buoyancy-driven interception. We evaluate ACT, Diffusion Policy, and the vision--language--action model SmolVLA on the platform. We investigate how learning methods and execution strategies developed for manipulation in air perform underwater, examining bubble disturbances, buoyancy-driven object motion, action-execution horizons, and real-time chunking. A separate single-arm study examines policy transfer between air and water and shows that demonstrations spanning both media support execution in both under the tested conditions. ULOHA provides a unified experimental platform for studying underwater bimanual robot learning under the coupled perceptual and physical effects of underwater environments. Additional material: https://mertcookimg.github.io/uloha/
Recent advances in vision-language-action models have stimulated growing interest in underwater embodied intelligence. However, their reliance on large-scale interaction data limits their applicability underwater, where data collection is costly and scarce. To address this challenge, we present AquaCap, a training-free Code-as-Policy framework for autonomous underwater navigation and manipulation. AquaCap employs a dual-layer agent that translates task instructions and environmental observations into condition-aware plans and executable control programs. Structured perception then provides the agent with semantic, geometric, and reliability-aware observations under degraded underwater conditions. A failure-aware memory diagnoses unsuccessful actions and supports closed-loop replanning and code revision. This design enables online adaptation without task-specific training or parameter updates. AquaCap achieves a 66.43% success rate in simulation. Real-world experiments further demonstrate autonomous grasping and object transport with an ROV, including the manipulation of targets displaced by hydrodynamic disturbances.
Xiaoshi Li, Yule Xu, Chunghiu Kong +5
Sun Yat-sen University · Dalian University of Technology · Southern Marine Science and Engineering Guangdong Laboratory
Underwater robotic grasping is difficult due to degraded, highly variable imagery and the expense of collecting diverse underwater demonstrations. We introduce a system that (i) autonomously collects successful underwater grasp demonstrations via a self-supervised data collection pipeline and (ii) transfers grasp knowledge from on-land human demonstrations through a depth-based affordance representation that bridges the on-land-to-underwater domain gap and is robust to lighting and color shift. An affordance model trained on on-land handheld demonstrations is deployed underwater zero-shot via geometric alignment, and an affordance-conditioned diffusion policy is then trained on underwater demonstrations to generate control actions. In pool experiments, our approach improves grasping performance and robustness to background shifts, and enables generalization to objects seen only in on-land data, outperforming RGB-only baselines. Code, videos, and additional results are available at https://umi-under-water.github.io.