NVIDIA Omniverse libraries provide modular APIs including ovrtx for RTX-based sensor simulation (camera, lidar, radar) from OpenUSD scenes, supporting real-time, physically grounded outputs for integration into existing application stacks. The lightweight C and Python SDK enables applications to configure render and sensor outputs, step simulation from the app loop, and retrieve output tensors on CPU or GPU.
Robots that can operate effectively in unpredictable, infrastructure-free environments face significant challenges, including reliability, environmental variability, and the need for real-world deployment to accumulate data. Most robotics companies fail because they optimize for controlled lab conditions rather than real-world scenarios. Burro Robotics, which operates in outdoor agricultural settings, has learned about tolerance and environmental variability through hands-on experience, and believes that a fundamentally different engineering achievement is needed to perform reliably across all conditions. Infrastructure-free localization and perception are key areas of research investment that will unlock the next phase of autonomous robotics for industrial environments.
BrainCo has developed a brain-controlled robot AI platform that allows users to direct a robot through neural signals in under 200 milliseconds. The technology works by decoding brain signals and converting them into commands for the robot, which can perform tasks such as grasping objects with precision. BrainCo's Embodied AI Data Collection Solution addresses the lack of high-quality training data for robots, capturing EEG data from human operators to record both physical actions and brain intent.
PyZeROS is an experimental Python-first alternative to `rclpy` for communicating with ROS 2, offering minimal dependencies, no ROS installation required, and asyncio executor. It can install through pip and communicates with ROS 2 through Zenoh, interoperating with standard ROS 2 nodes using rmw_zenoh_cpp. PyZeROS is designed for asyncio and asynchronous Python, providing custom ROS messages defined directly with Python classes.
OnSLAM is an open-source Windows application that runs LiDAR-inertial odometry and mapping on ROS1 bag files. It automatically detects compatible sensors, visualizes the trajectory and point-cloud map in a browser, and exports PLY and PCD files for easy inspection of datasets. The project aims to simplify the process of inspecting datasets without requiring a complete Linux and ROS environment setup. Users are encouraged to test the application with different sensors and provide feedback, particularly on Livox, Ouster, Velodyne, or Hesai data.
AGIBOT unveiled four new robots at WAIC 2026, expanding its portfolio of humanoid robots, industrial automation systems and dexterous robotic hands. The launches include the AGIBOT A3 Ultra humanoid robot, X2 EDU education platform, G2 Max industrial robot and OmniHand 3 Ultra-M dexterous robotic hand.
A real-time sEMG-based telecontrol system for an assistive robotic arm using a 1D convolutional neural network (CNN) achieves high classification performance, with a test accuracy above 90%, and stable real-time behavior. The system uses four-channel sEMG acquisition, signal preprocessing, segmentation into sliding windows, and classification. It is implemented in simulation and on a real robotic platform, producing coherent and smooth movements. The system demonstrates the feasibility of sEMG-based telecontrol for assistive robotics.
S.E.A.G.R, a socially and emotionally aware greeting robot framework, has been developed with dual-layer cultural and affective modulation. The system combines context-aware cultural mapping, emotion-based gesture modulation, and proxemic regulation in a unified Sense-Think-Act architecture. It aims to provide adaptive greetings for humans from diverse cultural backgrounds and different emotional states, addressing the limitations of existing static greeting routines.
PRISM, a multimodal perception system, leverages a custom sensor suite to capture RGB, depth, and thermal (RGB-D-T) imagery for terrain mapping in unstructured settings. At its core is OmniUnet, a novel vision transformer-based network designed for multimodal semantic terrain segmentation. PRISM efficiently generates traversability maps that enable autonomous navigation via a rover's Guidance, Navigation, and Control (GNC) subsystem.
A closed-form fifth-order transfer function is derived for the full controller-estimator-actuator-plant interconnection, and stability is characterized through the Routh-Hurwitz criterion over a parameterized linear model. Two representative three-parameter sweeps produce interpretable stability regions, and two uncertainty-aware tuning procedures are proposed: robustness-oriented design that maximizes gain margin and phase margin, and performance-oriented design that maximizes worst-case closed-loop bandwidth subject to margin constraints.
This paper proposes a safe reinforcement learning framework that learns control-affine dynamics with a certifiable data-driven safe policy using control barrier functions (CBF). It uses Control-Affine Random Fourier Features (ARFF) to model robot dynamics in a control-affine form and adaptive conformal prediction (ACP) to quantify the uncertainty in the safety constraint. The proposed framework demonstrates effectiveness on simulation results for the cartpole and 3D quadrotor platforms.
Foresight Residual RL for Long-Horizon Robot Manipulation with Vision-Language-Action Models optimizes handoff quality by augmenting subtask rewards with an offline-estimated foresight value, the probability of future subtask success conditioned on the terminal state of the current subtask. The method achieves 85.6% full-task success in a three-phase wrench-based nut-tightening assembly task, outperforming standard subtask residual RL and VLA baselines, while leaving per-subtask success unchanged.
Prashanth Chivkula and his co-authors developed a simulation platform that approximates the motion of a fish-like robot with computational efficiency. They applied PID control, where gains were learned using back propagation through time and training on a curriculum, to perform motion control and path tracking by the robot. The policy learned in the simulation was then applied on the physical platform, demonstrating an excellent match for agile fish-like robots.