Researchers at the University of Leeds and collaborators developed a 1.8 mm soft growing robot that can be steered magnetically, sense its own shape in real time, and operate without internal pressure, reducing friction and potentially improving patient outcomes following minimally invasive procedures. The robot uses magnetic particles embedded in silicone to create a "magnetic DNA" pattern that allows for both actuation and sensing, enabling real-time shape control and reduced risk of complications.
NVIDIA Open Sources First GPU-Accelerated Medical Physics Simulation Framework to help medical robotics developers model anatomy-device interaction, generate hard-to-capture scenarios, test in silico, and train or evaluate robot policies before hardware-heavy testing. The open-source framework brings together anatomy and medical device behavior with sensor simulation and robot learning to create reusable simulation environments, saving time and accelerating innovations to market faster.
ModPack is a modular and extensible teleoperation system designed to support diverse robot embodiments and task requirements within a unified framework. A wearable "backpack" integrates onboard computation, power, communication, and data storage at its core. The system supports plug-and-play capability modules including joint-level teleoperation with haptic feedback, mobile manipulation, and active perception. Experiments across two distinct robot platforms demonstrate that ModPack provides a flexible and reusable framework for data collection and policy learning.
Researchers have developed a drift-capable nonlinear model predictive control system to handle real-world winter driving conditions. The controller initiates and sustains drifting maneuvers to avoid collisions, achieving lower median lane error than electronic stability control at higher speeds.
Milo, the first open-source robotic guide dog platform, offers a fully autonomous solution for navigation assistance. The system consists of a modified robot equipped with onboard compute, sensors, and a handle, combined with a perception stack using voxel mapping and floor/obstacle/pedestrian detection. Milo is capable of navigating indoor and outdoor environments while avoiding obstacles and pedestrians, with a cost of approximately $2k USD compared to traditional guide dogs costing around $50k USD.
A framework called Learning from Haptics (LfH) is proposed to learn user-preferred safety interventions for haptic human-robot shared control. The framework uses a differentiable Control Barrier Function (CBF)-based optimization layer that adjusts safety parameters based on demonstrated haptic responses. Users teach the system how they expect it to intervene during teleoperation, resulting in haptic guidance reflecting their preferences while preserving intuitive interaction. Simulation and hardware experiments demonstrate the effectiveness of the proposed framework in learning personalized safety interventions from sparse user input.
LENS, or LLM-guided Environment Simplification, uses Large Language Models (LLMs) to automatically generate de-cluttered scene representations that improve classical planning and control in cluttered environments. By merging or pruning objects in a closed loop in response to task progress, LENS produces adaptive, task-relevant abstractions. Experiments show that LENS improves performance across various cluttered manipulation scenes using planning, model-based control, and vision-language-action models.
A decentralized, spoofing-aware trajectory planning framework for small unmanned aerial systems (UAS) has been presented. The proposed approach explicitly treats Remote Identification (RID) information as unverified and incorporates physical-layer observations to assess broadcast credibility. Received signal-strength measurements from neighboring aircraft are used to detect spoofing and probabilistically localize a spoofing agent, which is then converted into a risk-bounded unsafe region. This enables real-time, decentralized collision avoidance while preserving mission objectives and scalability.
NavVerse, a physics-enabled benchmark for indoor-to-outdoor embodied navigation, contains 100 indoor scenes, 50 urban outdoor scenes, and 50 indoor-to-outdoor scenes with 10,000 episodes across three tasks. Agents are evaluated through executable robot interfaces using task-success, path-efficiency, and safety metrics, revealing a clear drop in performance from outdoor to indoor-to-outdoor scenes, indicating adaptation as a major bottleneck. Zero-shot experiments show that end-to-end Vision-and-Language Assistants (VLAs) obtain the highest zero-shot success, while modular baselines provide the strongest safety profile.
A control-theoretic framework for contact-persistent full actuation in aerial physical interaction is introduced, combining a rigid-body model with a morphology-dependent wrench map. The main result shows that contact-persistent full actuation is equivalent to interiority of the task wrench in the constrained feasible wrench polytope and that the residual authority radius is exactly the distance to the polytope boundary.
A cable-driven limbless robot with rolling joints achieves versatile locomotion by reconfiguring body morphology and compliance, enabling diverse locomotion modes such as lateral undulation, sidewinding, rolling, and twisting. Distributed cable actuation generates traveling body waves, while programmable passive compliance enables robust contact-rich locomotion without terrain knowledge or high-bandwidth feedback. Rolling joints incorporate geared locking to maintain configuration without continuous power. Experiments demonstrate reliable gait generation, traversal in obstacle-rich environments, and transitions between modes.