DAISS: Phase-Aware Imitation Learning for Dual-Arm Robotic Ultrasound-Guided Interventions

Feng Li; Pei Liu; Shiting Wang; Ning Wang; Zhongliang Jiang; Nassir Navab; Yuan Bi

cs.ROMar 8, 2026

DAISS: Phase-Aware Imitation Learning for Dual-Arm Robotic Ultrasound-Guided Interventions

Authors: Feng Li, Pei Liu, Shiting Wang, Ning Wang, Zhongliang Jiang, Nassir Navab, Yuan Bi

Organizations: Chair for Computer Aided Medical Procedures(CAMP), Technical University of Munich(TUM), Munich, Germany · Munich Center for Machine Learning (MCML), Munich, Germany · The University of Hong Kong(HKU), Hong Kong, China

Abstract

Imitation learning has shown strong potential for automating complex robotic manipulation. In medical robotics, ultrasound-guided needle insertion demands precise bimanual coordination, as clinicians must simultaneously manipulate an ultrasound probe to maintain an optimal acoustic view while steering an interventional needle. Automating this asymmetric workflow -- and reliably transferring expert strategies to robots -- remains highly challenging. In this paper, we present the Dual-Arm Interventional Surgical System (DAISS), a teleoperated platform that collects high-fidelity dual-arm demonstrations and learns a phase-aware imitation policy for ultrasound-guided interventions. To avoid constraining the operator's natural behavior, DAISS uses a flexible NDI-based leader interface for teleoperating two coordinated follower arms. To support robust execution under real-time ultrasound feedback, we develop a lightweight, data-efficient imitation policy. Specifically, the policy incorporates a phase-aware architecture and a dynamic mask loss tailored to asymmetric bimanual control. Conditioned on a planned trajectory, the network fuses real-time ultrasound with external visual observations to generate smooth, coordinated dual-arm motions. Experimental results show that DAISS can learn personalized expert strategies from limited demonstrations. Overall, these findings highlight the promise of phase-aware imitation-learning-driven dual-arm robots for improving precision and reducing cognitive workload in image-guided interventions.

Figures & tables

Fig. 1: Overview of the DAISS Platform and Phase-Aware Imitation Learning. Our proposed bimanual teleoperation system integrates multimodal perception to execute complex ultrasound-guided interventions. By leveraging a dynamic phase-aware policy, DAISS effectively decouples the operation workflow, optimally balancing temporal efficiency with fine-grained kinematic precision.

Fig. 2: The DAISS Framework. (a) Leader Demonstration Generation: captures expert multimodal trajectories. (b) Robotic Follower Actuation: ensures safe, physical execution of the bimanual clinical workflow. (c) Phase-Aware Imitation Learning: maps the demonstrations to autonomous policies, employing a dynamic mask loss to resolve the speed-accuracy trade-off across different interventional phases.

Fig. 3: Phase-Aware Imitation Learning Module. Network module color scheme: multimodal inputs (blue), transformer-based action chunking (purple), phase-aware module (green), and dual-arm outputs (orange) controlling the FR3 (left) and Panda (right) manipulators.

Fig. 4: Kinematic collision avoidance . Green and yellow virtual cylinders represent the geometric envelopes of the robotic end-joints and tools. By acting as impenetrable spatial barriers, this setup inherently precludes mechanical interference between the dual manipulators during teleoperated maneuvers.

Fig. 5: Translational Tracking Performance .

Fig. 6: Rotational Tracking Performance .

Weight	0.1	0.5	1
Time(s)	$8.64\pm 1.15$	$12.44\pm 0.61$	$11.79\pm 0.70$
Weight	2	10
Time(s)	$15.18\pm 3.30$	$25.18\pm 4.52$

TABLE I: Temporal efficiency with different weights for phase II

Fig. 7: Kinematic trajectories across varying weights in Phase IV. The phase switch marker denotes the start of the final intervention stage.

Fig. 8: Phase II trajectory ablation . Kinematic execution of the anatomical scanning phase across varying weight parameters.

Explore similar work

EquiBim: Learning Symmetry-Equivariant Policy for Bimanual Manipulation
Mar 9, 2026Zhiyuan Zhang, Aditya Mohan, Seungho Han +3Bimanual Robotic Manipulation Robot Imitation Learning
Imitation Learning for Robot Assistance in Open Surgery: A Multi-Policy Evaluation on Suture Following
May 27, 2026Xucheng Wang, Zhizhou Yang, Xiaoman Zhang +3Visuomotor Policy Learning Robot Imitation Learning
Department of Biomedical Informatics, Harvard Medical School, Boston, MA · Department of Surgery, Massachusetts General Hospital, Boston, MA
Tri-Manual Visuomotor Imitation Learning of Robot Policies
Jul 28, 2026James Zhao, Mingyuan Ba, Weiming ZhiRobot Teleoperation Visuomotor Policy Learning
School of Computer Science and 2Australian Center For Robotics, The University of Sydney, Australia · College of Connected Computing, Vanderbilt University, TN, USA

Mar 9, 2026J/K move · Enter open · S save

EquiBim: Learning Symmetry-Equivariant Policy for Bimanual Manipulation

Zhiyuan Zhang, Aditya Mohan, Seungho Han, Wan Shou+2

Bimanual Robotic Manipulation Robot Imitation Learning Generalization in Robotic Manipulation

Open paper AskPDF arXiv

Robotic imitation learning has achieved impressive success in learning complex manipulation behaviors from demonstrations. However, many existing robot learning methods do not explicitly account for the physical symmetries of robotic systems, often resulting in asymmetric or inconsistent behaviors under symmetric observations. This limitation is particularly pronounced in dual-arm manipulation, where bilateral symmetry is inherent to both the robot morphology and the structure of many tasks. In this paper, we introduce EquiBim, a symmetry-equivariant policy learning framework for bimanual manipulation that enforces bilateral equivariance between observations and actions during training. Our approach formulates physical symmetry as a group action on both observation and action spaces, and imposes an equivariance constraint on policy predictions under symmetric transformations. The framework is model-agnostic and can be seamlessly integrated into a wide range of imitation learning pipelines with diverse observation modalities and action representations, including point cloud-based and image-based policies, as well as both end-effector-space and joint-space parameterizations. We evaluate EquiBim on RoboTwin, a dual-arm robotic platform with symmetric kinematics, and evaluate it across diverse observation and action configurations in simulation. We further validate the approach on a real-world dual-arm system. Across both simulation and physical experiments, our method consistently improves performance and robustness under distribution shifts. These results suggest that explicitly enforcing physical symmetry provides a simple yet effective inductive bias for bimanual robot learning.

Abstract

Explore similar work

EquiBim: Learning Symmetry-Equivariant Policy for Bimanual Manipulation

Imitation Learning for Robot Assistance in Open Surgery: A Multi-Policy Evaluation on Suture Following

Tri-Manual Visuomotor Imitation Learning of Robot Policies