Dr. Jianshu Zhou (NUS)

Robotic intelligence emerges not only from computation, but from the continuous interaction between the brain and the body. My research advocates a hand–brain co-development paradigm for embodied manipulation, in which task-level intelligence is jointly shaped by physical embodiment and cognitive decision-making. I view robotic hands and manipulators as embodied systems that actively structure perception and action through multimodal sensing, while higher-level intelligence provides reasoning, planning, and decision-making.

My prior research demonstrates how new grasping and manipulation mechanisms, novel actuation and transmission designs, and the integration of multimodal sensing, classical control, and data-driven policies can enable robots to perform tasks that were previously difficult or infeasible, and to execute manipulation with improved robustness, efficiency, and smoothness.

My research philosophy is to pursue work that is both intellectually interesting and practically useful, addressing fundamental questions while confronting real-world bottlenecks in robotics and embodied intelligence. My research path has been supported by several million USD in external investment, enabling the co-evolution of academic research and translational technology.

Representative Work 1: Fine In-Hand Manipulation

In this IEEE Transactions on Robotics (T-RO) 2024 work, we developed a dexterous robotic hand through innovative actuation, transmission, modeling, and control, enabling human-level performance in fine in-hand manipulation tasks. The system demonstrates a range of challenging behaviors, including in-hand plastic bag opening, in-hand object counting and sorting, cardboard box opening and card counting, as well as contact-rich bin picking and search tasks.

Paper Link: Jianshu Zhou, Junda Huang, Pieter Abbeel, Qi Dou, and Yunhui Liu. A Dexterous and Compliant (DexCo) Hand Based on Soft Hydraulic Actuation for Human Inspired Fine In-hand Manipulation. IEEE Transactions on Robotics, vol. 41, pp. 666-686, 2025, doi: 10.1109/TRO.2024.3508932.

In-hand plastic bag opening

The first robotic hand capable of opening a tightly closed plastic bag through fine in-hand manipulation.

Bagging objects

Putting objects into the opened plastic bag, and we are developing autonomous bag-tying capabilities.

Bagging objects

Adaptively screwing light bulbs on and off, with the ability to insert the bulb into the socket.

In-hand plastic bag opening

Counting a full stack of cards by grasping the front and bottom sides simultaneously using a human-inspired lateral cut motion.

Bagging objects

Cluttered bin picking that requires contact-rich manipulation to create space between crowded objects and insert fingers for the target grasp.

Bagging objects

Grasping multiple pills at once and sorting them into category-specific piles; we are developing an autonomous policy trained with multimodal visual–tactile information.

Representative Work 2: Anthropomorphic Dexterous Hand

This work introduces a one of the most dexterous five-fingered anthropomorphic robotic hand with 26 degrees of freedom, capable of achieving all 50 benchmarks of a healthy human hand. The hand serves as a comprehensive research platform for studying the relationship between kinematics, degree-of-freedom distribution, and resulting grasping and manipulation performance.

Paper Link: Jianshu Zhou, Xiaojiao Chen, Ukyoung Chang, Jui-Ting Lu, Clarisse Ching Yau Leung, Yonghua Chen, Zheng Wang. A soft-robotic approach to anthropomorphic robotic hand dexterity. IEEE Access, 7, pp. 2019. --Also see my IROS 2020.

X axis translation

Y axis translation

Z axis translation

X axis rotation

Y axis rotation

Z axis rotation

Representative Work 3: Universal Gripper (Cross-scale and Cross-state grasping)

Through an innovative Grasping–Suction Synergy, this work presents the first robotic gripper capable of achieving both cross-scale and cross-state grasping within a unified compliant architecture. The gripper robustly handles objects with surface areas ranging from 0.2 mm² to over 62,000 mm², spanning objects nearly 3500× smaller and 88× larger than its own contact area (~707 mm², 30 mm diameter). By integrating distributed suction, granular jamming, and tactile sensing, the system enables end-to-end grasping of real-world physical targets, including both solids and liquids, without requiring airtight sealing at the contact interface.

Paper Link: Jianshu Zhou, Jing Shu, Tianle Pan, Puchen Zhu, Jiajun An, Huayu Zhang, Upinder Kaur, Xin Ma, and Masayoshi Tomizuka, Everything-Grasping (EG) Gripper: A Universal Gripper with Synergistic Suction-Grasping Capabilities for Cross-Scale and Cross-State Manipulation, Soft Robotics (SORO). 2025.

0.5 mm glass bead grasping

A4 paper grasping

Unsealed surface woven bag grasping

Long metal bar grasping

Omnidirectional grasping

Packing of pills

Underwater grasping

Floating object grasping

Cross state grasping

Representative Work 4: Variable Stiffness Grasping

Contact remains a fundamental challenge in robotic grasping. To address this, we developed an innovative variable-stiffness phalange that mechanically adapts its stiffness to the applied grasping force, exhibiting an intrinsic form of mechanical intelligence. As the grasping force increases, the contact stiffness correspondingly increases; under lighter grasping forces, the finger remains compliant. This adaptive behavior enables robust contact regulation, enhances grasp stability, and allows the finger to withstand puncture from spiky objects, providing one of the earliest practical solutions for robotic durian grasping.

Paper Link: Jianshu Zhou, Yonghua Chen, Yong Hu, Zheng Wang, Yunquan Li, Guoying Gu, and Yunhui Liu. Adaptive variable stiffness particle phalange for robust and durable robotic grasping. Soft Robotics (SORO), 2020.

Representative Work 5: Origami Grasping

Extreme Compliant Grasping Enabled by a Scalable Non-Fingered Origami Gripper, capable of simultaneously grasping multiple delicate targets, including jellyfish, raw egg yolk, and rice.

Paper Link: Hanwen Cao, Jianshu Zhou, Kai Chen, Qiguang He, Qi Dou, and Yuihui Liu. Design and optimization of an origami gripper for versatile grasping and manipulation. Advanced Intelligent Systems (AIS), 2024. (Corresponding to Jianshu Zhou)

Jelly fish grasping

Raw egg yolk grasping and sorting

Grasping many at once

Representative Work 6: Modular Variable Stiffness Robot

We present the Programmable Locking Cell (PLC), a modular, tendon-driven unit that enables discrete and programmable stiffness modulation through mechanically interlocked joints. Each PLC transitions between compliant and rigid states via structural engagement, achieving up to 950% stiffness variation per unit without continuous power input. By assembling multiple PLCs, robotic structures with spatially programmable stiffness and reconfigurable morphology can be realized. The concept is validated through a variable-stiffness gripper for adaptive grasping and in-hand manipulation, and a pipe-traversing robot that combines shape adaptability with stiffness control in confined environments.

Paper Link: Jianshu Zhou, Wei Chen, Junda Huang, Boyuan Liang, Yunhui Liu, and Masayoshi Tomizuka. Programmable Locking Cells (PLC) for Modular Robots with High Stiffness Tunability and Morphological Adaptability. IEEE Transactions on Robotics (T-RO), 2025.

Representative Work 7: Variable Stiffness Robotic Wrist

Inspired by the human wrist, we developed a compact and lightweight soft wrist that enables hybrid motion and stiffness control through a novel multicable jamming mechanism. The wrist achieves agile orientation adjustment and task-adaptive stiffness modulation within a unified structure, supporting both compliant and forceful manipulation. Experiments and an automatic packaging task demonstrate its effectiveness in handling unknown objects in unstructured environments.

Paper Link: Jianshu Zhou, Hanwen Cao, Wei Chen, Shin Shing Cheng, and Yunhui Liu. Bioinspired soft wrist based on multicable jamming with hybrid motion and stiffness control for dexterous manipulation. IEEE/ASME Transactions on Mechatronics (T-Mech), 2022.

Representative Work 8: Tranformable and Reconfigurable Robot

We introduce the Prismatic–Bending Transformable (PBT) joint, a scissors-inspired modular mechanism that integrates bending, rotation, and extension/contraction within a single joint. The PBT joint enables transformable, reconfigurable, and scalable kinematic chains, allowing robotic manipulators to adapt their morphology to diverse tasks and constrained environments. Functioning as a single standardized module, PBT joints can be assembled into foldable manipulators or used as modular extensions for existing robotic systems. Experiments across multiple joint sizes demonstrate enhanced dexterity, reachability, and adaptability, particularly in confined and cluttered spaces.

Paper Link: Jianshu Zhou, Junda Huang, Boyuan Liang, Xiang Zhang, Xin Ma, and Masayoshi Tomizuka. Prismatic-Bending Transformable (PBT) Joint for a Modular, Foldable Manipulator with Enhanced Reachability and Dexterity. IEEE/ASME Transactions on Mechatronics (T-Mech), 2025.

Representative Work 9: Behavior Cloning of Fine In-hand Manipulation

We present DIH-Tele, a dexterous in-hand teleoperation framework that integrates a tactile dexterous hand, accurate teleoperation, multimodal data collection, and an imitation learning algorithm to enable complex multi-object manipulation. Using in-hand counting and selective removal as a representative task, the system leverages multimodal perception to achieve performance approaching that of human teleoperation. Experimental results and ablation studies demonstrate the importance of multimodal sensing and fused training in learning versatile, high-dimensional fingertip behaviors.

Paper Link: Paper Link: Junda Huang, Kai Chen, Jianshu Zhou, Xinyu Lin, Pieter Abbeel, Qi Dou, and Yuhui Liu. DIH-Tele: Dexterous In-Hand Teleoperation Framework for Learning Multi-Objects Manipulation With Tactile Sensing. IEEE/ASME Transactions on Mechatronics (T-Mech), 2025. (Corresponding to Jianshu Zhou)

Learning from Demonstration via Teleoperation for In-Hand Multi-Object Counting and Sorting

Representative Work 10: Innovative Teleoperation Framework

We propose a Global–Local teleoperation interface that decouples robot control into a global component for large-range motion and collision-aware navigation, and a local component for precise, dexterous manipulation. This design unifies workspace coverage and fine motor control, enabling efficient teleoperation across tasks ranging from long-range manipulation to delicate, contact-rich operations with dexterous end-effectors. Extensive experiments demonstrate improved accuracy, usability, and generalizability, while also providing a scalable interface for high-quality demonstration data collection.

Paper Link: Jianshu Zhou, Boyuan Liang, Junda Huang, Ian Zhang, Pieter Abbeel, Masayoshi Tomizuka. Global-local interface for on-demand teleoperation. arXiv e-prints. 2025 Feb:arXiv-2502.

Opening an iPhone SIM slot using a pin

Key insertion and lock open