Accurate and robust gait phase estimation is crucial for exoskeleton control in assistive walking. Existing data-driven techniques, while powerful, lack the predictability and generalizability needed for real-world use. Model-based methods, like adaptive oscillators (AOs), offer inherent certainty but cannot adapt to varying terrains. This study introduces a data-augmented adaptive oscillator (DA-AO), a hybrid framework synergizing mathematical models with the adaptability of learning-based perception. By dynamically retrieving the optimal hyperparameters for each detected locomotion mode, the DA-AO ensures that the oscillator consistently operates in its optimal regime, thereby achieving rapid adaptation across diverse walking scenarios. The proposed estimator is validated using a hip exoskeleton, achieving highly accurate estimation with root mean square errors (RMSEs) of 0.025, 0.021, and 0.032 for level-ground walking, ramp ascent, and stair ascent, respectively, during active assistive walking trials that effectively reduced average metabolic costs by 15.3%, 10.6%, and 5.7% ($p< 0.05$). Furthermore, the DA-AO system demonstrated robust gait phase estimation with an RMSE of 0.025 during continuous outdoor terrain transitions and reduced the cost of transport by 12.3% ($p< 0.05$) during outdoor hill walking. This work establishes a practical integration of data-driven adaptation within a model-based framework, demonstrating the potential of hybrid strategies to resolve the accuracy-reliability trade-off and foster user trust in assistive walking with exoskeletons.
This paper presents a distributed cooperative-planning framework for heterogeneous robots operating in shared and space-constrained environments. We propose novel geometry-aware collision-free constraints that support low-conservatism representations of heterogeneous robot shapes. These constraints are analytical and separable, making them naturally suited for distributed optimization. Leveraging this separable nature, we develop a distributed planning framework based on the alternating direction method of multipliers that decomposes the computational burden across robots. The framework directly handles nonlinear and non-convex coupling constraints without linearization, and any converged solution satisfies the Karush-Kuhn-Tucker conditions of the original global cooperative-planning problem. To improve per-robot computational efficiency, acceleration techniques including dimension reduction and block coordinate descent are incorporated. Moreover, clear geometric interpretations of the coordination mechanism are provided to enhance algorithmic interpretability in practical deployment. Comprehensive validation through simulations, hardware experiments, and benchmark comparisons demonstrates the framework's effectiveness in enabling safe, high-quality, and efficient coordination for heterogeneous robots.
Impedance control is a widely adopted strategy for robotic systems physically interacting with their environments. However, it is well known that impedance control is vulnerable to disturbances and parametric uncertainties. This paper proposes a control design to impart robustness against disturbances and parametric uncertainties to impedance control, in the sense of nonlinear $\mathcal {H}_{\infty }$ optimality—namely, nonlinear $\mathcal {H}_{\infty }$ impedance control. An analytic solution to the nonlinear $\mathcal {H}_{\infty }$ inverse-optimal control problem is derived for the interconnected system of the robot and its environment. The resulting optimal control takes the form of reference error feedback. Robust stability is guaranteed by extended disturbance input-to-state stability. Passivity of the interconnected robot–environment system is ensured, thereby realizing robust and passive impedance control. Experimental validations demonstrate that the proposed $\mathcal {H}_{\infty }$ impedance control achieves robust stability and performance in the presence of disturbances and parametric uncertainties.
Soft parallel robots with their manipulation safety and low commercial cost show a promising future for delicate operations and safe human-robot interactions. However, promoting the use of electroactive polymers (EAPs) is still challenging due to the complexity of modelling the multi-actuator collaboration. Taking the viscoelasticity of the material into account, we developed an inverse dynamic model for the collaboration of multiple dielectric elastomer actuators (DEAs). Two DEA-based parallel kinematics robots were designed and fabricated to validate the model. A generic method for fabricating a high-stability conductive paint on a silicon-based film has been achieved by sandwiching photosensitive resin mixed with conductive particles on laser engraved DE-film. Compared to the widely used low-cost carbon grease, the fabricated electrode shows a higher consistency in its dynamic behaviour before and after a period of static resting. The validation results show a robot output force prediction with 12.1% RMSE under static end-effector conditions, and a trajectory with 3.2% RMSE overall.
Neural fields (NFs) and 3D Gaussian Splatting (3DGS) have emerged as a transformative approach for 3D scene representation in computer vision and robotics, enabling accurate inference of geometry, 3D semantics, and dynamics from posed 2D data. NFs and 3DGS encompass both continuous implicit and explicit 3D representations, enabling high-fidelity 3D reconstruction, integration of multi-modal sensor data, and generation of novel viewpoints. This survey explores their applications in robotics, emphasizing their potential to enhance perception, planning, and control. Their compactness, memory efficiency, and differentiability make them ideal for real-time robotics applications. In addition, their seamless integration with foundation and generative models can improve robot adaptability and decision-making. This paper provides a thorough review of NFs and 3DGS in robotics, categorizing applications across various domains and evaluating their strengths and limitations, based on over 250 papers. First, we present five key frameworks: Occupancy Networks, Signed Distance Fields, Neural Radiance Fields, 3D Gaussian Splatting and Feature Fields. Second, we detail applications in five major robotics domains: pose estimation, manipulation, navigation, physics, and autonomous driving; highlighting key works and discussing takeaways and open challenges. Finally, we outline the current limitations of NFs and 3DGS in robotics and propose promising directions for future research. Project page: robonerf.github.io
