Demonstrators
ROBOCOMPLEX
ROBOCOMPLEX
Navigation through the Somport Tunnel .
RoboBoat: A robotic boat for 3D mapping of partially flooded underground sites.
Planning and navigation in dynamic environments (University hall).
Tunnels, open-pit mines, roads under construction (G-Loc, LG-SLAM).
Autonomous bathymetry and geopointing to enhance communications (Pantano de las Navas).
Autonomous inspection of water channels with USV (Canal-túnel de Estada).
We present "DWA-3D" an extension of the well-known "Dynamic Window Approach" reactive planner to the 3D space and implement it for the first time in a real system. The integrated architecture represents a proof of concept to perform autonomous UAV navigation in indoor or confined scenarios, where a safe and high maneuverability is required. The system has faced several real scenarios, validating its capabilities to avoid collisions. The autonomous UAV is able to perform safe inspection missions even in poorly light conditions or dusty environments as it relies on an onboard 3D-LiDAR both for navigation and self-localization.
The video presents a brief explanation of our high-speed flight system in tunnels for UAVs. By using two LiDAR scanners and a neural network, the system predicts the drone's orientation relative to the tunnel's axis and allows for yaw correction to keep both aligned. This enables the drone to fly at speeds of up to 10 m/s while avoiding collisions, making it viable for use in large underground environments. Additionally, the video showcases preliminary experiments combining a real platform with simulated data to validate the system.
Video presentation of the work submitted to the International Conference on Intelligent Robots and Systems (IROS 2024). In it, we introduce a procedural tunnel network generation system that enables the validation and training of robotic systems designed for underground environments. The video provides a brief explanation of how the system works, displays the results obtained, and highlights the use cases of these procedurally generated environments in other research projects.
Roboboat is a robotized boat with a new autonomous navigation contour-following navigation technique, the Sliding Balloon, which combines a simple planner for computing the next subgoal for following the uneven contour, avoiding trap situations, and a reactive technique to safety move towards this goal, taking into account the kinodynamic boat constraints. An ICP-based SLAM method has been developed adapted for use in unstructured confined environments, which have highly irregular contours and walls. The localization and mapping are directly based on the point clouds generated from two different LiDAR sensors working at different sampling periods. The reconstruction method is performed offline, in two consecutive steps. In the first step, the USV trajectory is computed after filtering the point clouds to remove unwanted effects, In the second step, this trajectory is used to compose the final 3D map by merging the point clouds provided by the two LiDARs after downsampling.
The Traversability-aware FMM path planner avoids regions with moving obstacles that pose collision risks. It divides the environment into regions, assigning traversability values based on obstacle movement and path deviation.
MADA (Monitoring with Avoidance of Dynamic Areas) is a planner for monitoring environments with dynamic obstacles. It selects observation goals based on environmental distribution and monitoring costs, then generates paths that avoid densely occupied dynamic regions.
An occupation-aware planning method for multi-robot teams in dynamic environments. Robots share observed data, select monitoring regions, and compute paths that avoid areas with high obstacle density.
AVOCADO: Adaptive Optimal Collision Avoidance: Driven by Opiniona novel navigation approach to address holonomic robot collision avoidance when the robot does not know how cooperative the other agents in the environment are. AVOCADO departs from a velocity obstacle’s (VO) formulation akin to the optimal reciprocal collision avoidance method. However, instead of assuming reciprocity, it poses an adaptive control problem to adapt to the cooperation level of other robots and agents in real time. This is achieved through a novel nonlinear opinion dynamics design that relies solely on sensor observations. As a by-product, we leverage tools from the opinion dynamics formulation to naturally avoid the deadlocks in geometrically symmetric scenarios that typically sufferVO-based planners
RUMOR (Reinforcement learning for Understanding a MOdel of the Real world), a motion planner that combines model-based and DRL benefits. It uses a deep abstraction of the environment provided by the Dynamic Object Velocity Space (DOVS) to understand the surrounding scenario. The DOVS model represents the dynamism and the future of the environment in the velocity space of the robot. The planner applies this information as an input of the DRL algorithm, learning to interpret the future dynamic sce nario knowledge, taking advantage over other approaches that use raw obstacle information and are not able to generalize in scenarios that are different from those previously experienced.
Robot navigation methods allow mobile robots to operate in applications such as warehouses or hospitals. While the environment in which the robot operates imposes requirements on its navigation behavior, most existing methods do not allow the end-user to configure the robot's behavior and priorities, possibly leading to undesirable behavior (e.g., fast driving in a hospital). We propose a novel approach to adapt robot motion behavior based on natural language instructions provided by the end-user. Our zero-shot method uses an existing Visual Language Model to interpret a user text query or an image of the environment. This information is used to generate the cost function and reconfigure the parameters of a Model Predictive Controller, translating the user's instruction to the robot's motion behavior. This allows our method to safely and effectively navigate in dynamic and challenging environments.
G-Loc: Tightly-coupled Graph Localization with Prior Topo-metric Information: This video showcases the performance of the G-Loc algorithm using data from the EULT dataset, as well as data collected on the University of Zaragoza campus using a sensor-equipped vehicle. G-Loc is a map-based localization framework that seamlessly integrates LiDAR, inertial and GNSS measurements with cloud-to-map registration in a sliding window graph fashion. G-Loc offers a new perspective on map-based localization by reusing prior topological and metric information. The video illustrates how the sliding graph formed by recent robot poses (whose nodes are shown as blue squares and edges as blue lines) is jointly optimized with a previously built topo-metric map, shown in orange.
LG-SLAM system performance in different scenarios is shown in these video. Sequences are recorded with an Ouster with 128-planes and an Alphasense IMU carried hand-held.
Water, a vital but scarce resource, has necessitated the development of extensive distribution and storage infrastructure. Given its limited availability, minimizing losses from poorly maintained structures is imperative. The age and deteriorated condition of some infrastructure pose significant challenges and risks for human inspectors. One of our research lines addresses this by focusing on the autonomous navigation of Unmanned Surface Vehicles (USVs) in diverse environments, facilitating tasks such as centimeter-accurate lake bathymetry and water-channel inspection for digital-twin reconstruction.
The video shows two key tasks: autonomous tunnel mapping using a robotic boat, and maintaining reliable communications with directional antennas to ensure stable connectivity in challenging environments.