Accurate driving force estimation is critical for ground vehicles operating in soft-soil terrains, where complex wheel–soil interactions involving sinkage and shear effects significantly affect vehicle mobility. Conventional Driving Force Observers (DFOs), initially developed for rigid road conditions, fail to account for these terrain-specific nonlinearities, leading to estimation offset and substantial errors. To overcome this limitation, this paper proposes a Terramechanics-Enhanced Wheel Dynamics (TEWD) model that explicitly integrates soil resistance derived from established wheel–soil interaction models. Based on the TEWD model, we develop a new driving force observer, termed TEWD-DFO, that estimates driving forces solely from internal vehicle signals without requiring additional external force sensors. The proposed method is validated through simulations conducted in the ProjectChrono environment under soft-soil conditions. Simulation results demonstrate that the TEWD-DFO significantly reduces estimation offsets and achieves robust driving force estimation performance, indicating its potential applicability in enhancing vehicle stability and control accuracy in realistic soft-soil scenarios.

We present the Keras domain packages KerasCV and KerasNLP, extensions of the Keras API for Computer Vision and Natural Language Processing workflows, capable of running on either JAX, TensorFlow, or PyTorch. These domain packages are designed to enable fast experimentation, with a focus on ease-of-use and performance. We adopt a modular, layered design: at the library's lowest level of abstraction, we provide building blocks for creating models and data preprocessing pipelines, and at the library's highest level of abstraction, we provide pretrained "task" models for popular architectures such as Stable Diffusion, YOLOv8, GPT2, BERT, Mistral, CLIP, Gemma, T5, etc. Task models have built-in preprocessing, pretrained weights, and can be fine-tuned on raw inputs. To enable efficient training, we support XLA compilation for all models, and run all preprocessing via a compiled graph of TensorFlow operations using the tf.data API. The libraries are fully open-source (Apache 2.0 license) and available on GitHub. Keywords: KerasCV, KerasNLP, Keras multi-backend, Deep learning, Generative AI

Differential elastic actuators connect a motor and a spring via differential gears to a shared output shaft, offering a more compact solution for creating mechanically robust systems than series elastic actuators, with superior torque transmission at high frequencies. The key to maximizing DEA performance lies in the careful selection of stiffness, inertia, and damping values to meet specific requirements for performance and durability. We introduce a DEA design guide that utilizes open-loop torque-bandwidth for performance evaluation and the magnitude of impact-induced gear torque for robustness evaluation. This approach enables determining DEA parameters using closed-form equations, eliminating the need for simulations or extensive expert knowledge. The effectiveness of our method is confirmed through experiments with a reconfigurable DEA prototype.

SwarmRail represents a novel solution to overhead manipulation from a mobile unit that drives in an aboveground rail-structure. The concept is based on the combination of omnidirectional mobile platform and L-shaped rail profiles that form a through-going central gap. This gap makes possible mounting a robotic manipulator arm overhead at the underside of the mobile platform. Compared to existing solutions, SwarmRail enables continuous overhead manipulation while traversing rail crossings. It also can be operated in a robot swarm, as it allows for concurrent operation of a group of mobile SwarmRail units inside a single rail network. Experiments on a first functional demonstrator confirm the functional capability of the concept. Potential fields of applications reach from industry over logistics to vertical farming.

In this article we present a sagittal plane, sprawled posture hexapedal running model with distributed body inertia, massless legs and serial elastic actuation at the hips as well as along the telescoping legs. We show by simulation that simple, periodic, feedforward controlled actuation is sufficient to obtain steady period 1 running gaits at twice the actuation frequency. We observe a nearly linear relation of average running speed and actuation frequency. The ground reaction profiles of the legs show leg specialization as observed in running insects. Interleg phasing has a strong influence on the foot fall sequence and thus the overall body dynamics. While the single leg ground reaction force profiles show little dependency on interleg actuation phase the total reaction force does. Thus, depending on the interleg actuation phase body motions without flight phase are observed as well as body motions and total ground reaction forces that show similarities to those obtained for the spring loaded inverted pendulum model. Further, we show that including leg damping and a ground friction model the periodic orbits have a large region of attraction with respect to the initial conditions. Additionally, the model quickly rejects step up and step down disturbances as well as force impulses. Finally, we briefly discuss the energetics of the hexapedal running model.

Compliant actuators in robotic systems improve robustness against rigid impacts and increase the performance and efficiency of periodic motions such as hitting, jumping and running. However, in the case of rigid impacts, as they can occur during hitting or running, the system behavior is changed compared to free motions which turns the control into a challenging task. We introduce a controller that excites periodic motions along the direction of an intrinsic mechanical oscillation mode. The controller requires no model knowledge and adapts to a modal excitation by means of measurement of the states. We experimentally show that the controller is able to stabilize a hitting motion on the variable stiffness robot DLR Hand Arm System. Further, we demonstrate by simulation that the approach applies for legged robotic systems with compliantly actuated joints. The controlled system can approach different modes of motion such as jumping, hopping and running, and thereby, it is able to handle the repeated occurrence of robot-ground contacts.

For autonomous navigation tasks it is important that the robot always has a good estimate of its current pose with respect to its starting position and - in terms of orientation - with respect to the gravity vector. For this, the robot should make use of all available information and be robust against the failure of single sensors. In this paper a multisensor data fusion algorithm for the six-legged walking robot DLR Crawler is presented. The algorithm is based on an indirect feedback information filter that fuses measurements from an inertial measurement unit (IMU) with relative 3D leg odometry measurements and relative 3D visual odometry measurements from a stereo camera. Errors of the visual odometry are computed and considered in the filtering process in order to achieve accurate pose estimates which are robust against visual odometry failure. The algorithm was successfully tested and results are presented.

This article analyzes and evaluates the stability of the biologically inspired gait of the DLR Crawler, a walking hexapod robot, with respect to leg loss. Using a kinematic simulation, ranges of velocity commands that result in stable gait coordination are determined for both cases, the undamaged robot and the robot experiencing the loss of a single leg. The results give insight how to adjust the motion commands after the loss of a leg. Further, a simplified dynamic simulation is used to analyze the effect of leg loss on the walking stability. Heuristic measures like curvature and length of the traveled path, roll and pitch angles are employed to evaluate the walking stability and performance. Some methods like shifting the COG or stiffening the variably compliant joints are proposed and discussed with respect to their ability to improve the walking performance in case of leg loss. In the end, the presented concepts are extended and for the first time applied to a simulated eight-legged robot.

Walking is a fascinating way of locomotion that is very robust, especially in unstructured terrain. Many researchers devote their time to understanding its underlying principles and to build robots based on their findings. Using the fingers of DLR-Hand II a six-legged actively compliant walking robot is developed. It is intended to be used as testbed for the evaluation of different force- and position-based leg and gait control algorithms for hexapod walking in rough terrain. Following a brief overview of the finger hardware, the use of fingers as legs is analyzed and discussed. The body geometry as well as the systems constituting the robot are described. The compliance control algorithm used is explained and finally some experimental results are presented.