Learning robust autonomous navigation and locomotion for wheeled-legged robots

Autonomous wheeled-legged robots have the potential to transform logistics systems, improving operational efficiency and adaptability in urban environments. Navigating urban environments, however, poses unique challenges for robots, necessitating innovative solutions for locomotion and navigation. These challenges include the need for adaptive locomotion across varied terrains and the ability to navigate efficiently around complex dynamic obstacles. This work introduces a fully integrated system comprising adaptive locomotion control, mobility-aware local navigation planning, and large-scale path planning within the city. Using model-free reinforcement learning (RL) techniques and privileged learning, we developed a versatile locomotion controller. This controller achieves efficient and robust locomotion over various rough terrains, facilitated by smooth transitions between walking and driving modes. It is tightly integrated with a learned navigation controller through a hierarchical RL framework, enabling effective navigation through challenging terrain and various obstacles at high speed. Our controllers are integrated into a large-scale urban navigation system and validated by autonomous, kilometer-scale navigation missions conducted in Zurich, Switzerland, and Seville, Spain. These missions demonstrate the system's robustness and adaptability, underscoring the importance of integrated control systems in achieving seamless navigation in complex environments. Our findings support the feasibility of wheeled-legged robots and hierarchical RL for autonomous navigation, with implications for last-mile delivery and beyond.

ANYmal parkour: Learning agile navigation for quadrupedal robots

Performing agile navigation with four-legged robots is a challenging task because of the highly dynamic motions, contacts with various parts of the robot, and the limited field of view of the perception sensors. Here, we propose a fully learned approach to training such robots and conquer scenarios that are reminiscent of parkour challenges. The method involves training advanced locomotion skills for several types of obstacles, such as walking, jumping, climbing, and crouching, and then using a high-level policy to select and control those skills across the terrain. Thanks to our hierarchical formulation, the navigation policy is aware of the capabilities of each skill, and it will adapt its behavior depending on the scenario at hand. In addition, a perception module was trained to reconstruct obstacles from highly occluded and noisy sensory data and endows the pipeline with scene understanding. Compared with previous attempts, our method can plan a path for challenging scenarios without expert demonstration, offline computation, a priori knowledge of the environment, or taking contacts explicitly into account. Although these modules were trained from simulated data only, our real-world experiments demonstrate successful transfer on hardware, where the robot navigated and crossed consecutive challenging obstacles with speeds of up to 2 meters per second.

Reinforcement and Curriculum Learning for Off-Road Navigation of an UGV with a 3D LiDAR

This paper presents the use of deep Reinforcement Learning (RL) for autonomous navigation of an Unmanned Ground Vehicle (UGV) with an onboard three-dimensional (3D) Light Detection and Ranging (LiDAR) sensor in off-road environments. For training, both the robotic simulator Gazebo and the Curriculum Learning paradigm are applied. Furthermore, an Actor-Critic Neural Network (NN) scheme is chosen with a suitable state and a custom reward function. To employ the 3D LiDAR data as part of the input state of the NNs, a virtual two-dimensional (2D) traversability scanner is developed. The resulting Actor NN has been successfully tested in both real and simulated experiments and favorably compared with a previous reactive navigation approach on the same UGV.

Automatically Annotated Dataset of a Ground Mobile Robot in Natural Environments via Gazebo Simulations

This paper presents a new synthetic dataset obtained from Gazebo simulations of an Unmanned Ground Vehicle (UGV) moving on different natural environments. To this end, a Husky mobile robot equipped with a tridimensional (3D) Light Detection and Ranging (LiDAR) sensor, a stereo camera, a Global Navigation Satellite System (GNSS) receiver, an Inertial Measurement Unit (IMU) and wheel tachometers has followed several paths using the Robot Operating System (ROS). Both points from LiDAR scans and pixels from camera images, have been automatically labeled into their corresponding object class. For this purpose, unique reflectivity values and flat colors have been assigned to each object present in the modeled environments. As a result, a public dataset, which also includes 3D pose ground-truth, is provided as ROS bag files and as human-readable data. Potential applications include supervised learning and benchmarking for UGV navigation on natural environments. Moreover, to allow researchers to easily modify the dataset or to directly use the simulations, the required code has also been released.

A Simulated Environment for Robot Vision Experiments

Training on simulation data has proven invaluable in applying machine learning in robotics. However, when looking at robot vision in particular, simulated images cannot be directly used no matter how realistic the image rendering is, as many physical parameters (temperature, humidity, wear-and-tear in time) vary and affect texture and lighting in ways that cannot be encoded in the simulation. In this article we propose a different approach for extracting value from simulated environments: although neither of the trained models can be used nor are any evaluation scores expected to be the same on simulated and physical data, the conclusions drawn from simulated experiments might be valid. If this is the case, then simulated environments can be used in early-stage experimentation with different network architectures and features. This will expedite the early development phase before moving to (harder to conduct) physical experiments in order to evaluate the most promising approaches. In order to test this idea we created two simulated environments for the Unity engine, acquired simulated visual datasets, and used them to reproduce experiments originally carried out in a physical environment. The comparison of the conclusions drawn in the physical and the simulated experiments is promising regarding the validity of our approach.

Learning quadrupedal locomotion over challenging terrain

Legged locomotion can extend the operational domain of robots to some of the most challenging environments on Earth. However, conventional controllers for legged locomotion are based on elaborate state machines that explicitly trigger the execution of motion primitives and reflexes. These designs have increased in complexity but fallen short of the generality and robustness of animal locomotion. Here, we present a robust controller for blind quadrupedal locomotion in challenging natural environments. Our approach incorporates proprioceptive feedback in locomotion control and demonstrates zero-shot generalization from simulation to natural environments. The controller is trained by reinforcement learning in simulation. The controller is driven by a neural network policy that acts on a stream of proprioceptive signals. The controller retains its robustness under conditions that were never encountered during training: deformable terrains such as mud and snow, dynamic footholds such as rubble, and overground impediments such as thick vegetation and gushing water. The presented work indicates that robust locomotion in natural environments can be achieved by training in simple domains.

Learning-Based Methods of Perception and Navigation for Ground Vehicles in Unstructured Environments: A Review

The problem of autonomous navigation of a ground vehicle in unstructured environments is both challenging and crucial for the deployment of this type of vehicle in real-world applications. Several well-established communities in robotics research deal with these scenarios such as search and rescue robotics, planetary exploration, and agricultural robotics. Perception plays a crucial role in this context, since it provides the necessary information to make the vehicle aware of its own status and its surrounding environment. We present a review on the recent contributions in the robotics literature adopting learning-based methods to solve the problem of environment perception and interpretation with the final aim of the autonomous context-aware navigation of ground vehicles in unstructured environments. To the best of our knowledge, this is the first work providing such a review in this context.

Reactive Navigation on Natural Environments by Continuous Classification of Ground Traversability

Reactivity is a key component for autonomous vehicles navigating on natural terrains in order to safely avoid unknown obstacles. To this end, it is necessary to continuously assess traversability by processing on-board sensor data. This paper describes the case study of mobile robot Andabata that classifies traversable points from 3D laser scans acquired in motion of its vicinity to build 2D local traversability maps. Realistic robotic simulations with Gazebo were employed to appropriately adjust reactive behaviors. As a result, successful navigation tests with Andabata using the robot operating system (ROS) were performed on natural environments at low speeds.

Regressed Terrain Traversability Cost for Autonomous Navigation Based on Image Textures

The exploration of remote, unknown, rough environments by autonomous robots strongly depends on the ability of the on-board system to build an accurate predictor of terrain traversability. Terrain traversability prediction can be made more cost efficient by using texture information of 2D images obtained by a monocular camera. In cases where the robot is required to operate on a variety of terrains, it is important to consider that terrains sometimes contain spiky objects that appear as non-uniform in the texture of terrain images. This paper presents an approach to estimate the terrain traversability cost based on terrain non-uniformity detection (TNUD). Terrain images undergo a multiscale analysis to determine whether a terrain is uniform or non-uniform. Terrains are represented using a texture and a motion feature computed from terrain images and acceleration signal, respectively. Both features are then combined to learn independent Gaussian Process (GP) predictors, and consequently, predict vibrations using only image texture features. The proposed approach outperforms conventional methods relying only on image features without utilizing TNUD.

Supervised Learning of Natural-Terrain Traversability with Synthetic 3D Laser Scans

Autonomous navigation of ground vehicles on natural environments requires looking for traversable terrain continuously. This paper develops traversability classifiers for the three-dimensional (3D) point clouds acquired by the mobile robot Andabata on non-slippery solid ground. To this end, different supervised learning techniques from the Python library Scikit-learn are employed. Training and validation are performed with synthetic 3D laser scans that were labelled point by point automatically with the robotic simulator Gazebo. Good prediction results are obtained for most of the developed classifiers, which have also been tested successfully on real 3D laser scans acquired by Andabata in motion.