This paper introduces a novel proprioceptive state estimator for legged robots that combines model-based filters with deep neural networks. In environments where vision systems are not reliable, proprioceptive state estimators become indispensable. Traditionally, proprioceptive state estimators are based on model-based approaches, which rely solely on contact foot kinematics as measurements. In contrast, learning-based approaches have obtained new measurements, such as displacement and covariance, by leveraging real-world data in a supervised manner. In this work, we develop a state estimation framework that trains a neural measurement network (NMN) to estimate the base's linear velocity and foot contact probability, which are then employed as measurements in an invariant extended Kalman filter. Our approach relies solely on simulation data for training, as it allows us to obtain extensive data easily. We address the sim-to-real gap by adapting existing learning techniques and regularization. To validate our proposed method, we conduct hardware experiments using a quadruped robot on four types of terrain: flat, debris, soft, and slippery. In our experiments, the proposed method demonstrates significant improvements over the model-based state estimator, achieving an average reduction in Absolute Trajectory Error (ATE) by${6 1. 8 \%}$for position and${8. 5 \%}$for velocity.

We present ArtiGrasp, a novel method to synthesize bimanual hand-object interactions that include grasping and articulation. This task is challenging due to the diversity of the global wrist motions and the precise finger control that are necessary to articulate objects. ArtiGrasp leverages reinforcement learning and physics simulations to train a policy that controls the global and local hand pose. Our framework unifies grasping and articulation within a single policy guided by a single hand pose reference. Moreover, to facilitate the training of the precise finger control required for articulation, we present a learning curriculum with increasing difficulty. It starts with single-hand manipulation of stationary objects and continues with multi-agent training including both hands and non-stationary objects. To evaluate our method, we introduce Dynamic Object Grasping and Articulation, a task that involves bringing an object into a target articulated pose. This task requires grasping, relocation, and articulation. We show our method’s efficacy towards this task. We further demonstrate that our method can generate motions with noisy hand-object pose estimates from an off-the-shelf image-based regressor. Project page: https://eth-ait.github.io/artigrasp/.

Several earlier studies have shown impressive control performance in complex robotic systems by designing the controller using a neural network and training it with model-free reinforcement learning. However, these outstanding controllers with natural motion style and high task performance are developed through extensive reward engineering, which is a highly laborious and time-consuming process of designing numerous reward terms and determining suitable reward coefficients. In this work, we propose a novel reinforcement learning framework for training neural network controllers for complex robotic systems consisting of bothrewardsandconstraints. To let the engineers appropriately reflect their intent to constraints and handle them with minimal computation overhead, two constraint types and an efficient policy optimization algorithm are suggested. The learning framework is applied to train locomotion controllers for several legged robots with different morphology and physical attributes to traverse challenging terrains. Extensive simulation and real-world experiments demonstrate that performant controllers can be trained with significantly less reward engineering, by tuning only a single reward coefficient. Furthermore, a more straightforward and intuitive engineering process can be utilized, thanks to the interpretability and generalizability of constraints. The summary video is available athttps://youtu.be/KAlm3yskhvM.

We introduce the dynamic grasp synthesis task: given an object with a known 6D pose and a grasp reference, our goal is to generate motions that move the object to a target 6D pose. This is challenging, because it requires reasoning about the complex articulation of the human hand and the intricate physical interaction with the object. We propose a novel method that frames this problem in the reinforcement learning framework and leverages a physics simulation, both to learn and to evaluate such dynamic interactions. A hierarchical approach decomposes the task into low-level grasping and high-level motion synthesis. It can be used to generate novel hand sequences that approach, grasp, and move an object to a desired location, while retaining human-likeness. We show that our approach leads to stable grasps and generates a wide range of motions. Furthermore, even imperfect labels can be corrected by our method to generate dynamic interaction sequences. Video and code are available at: https://eth-ait.github.io/d-grasp/.

In this work, a non-gaited framework for legged system locomotion is presented. The approach decouples the gait sequence optimization by considering the problem as a decision-making process. The redefined contact sequence problem is solved by utilizing a Monte Carlo Tree Search (MCTS) algorithm that exploits optimization-based simulations to evaluate the best search direction. The proposed scheme has proven to have a good trade-off between exploration and exploitation of the search space compared to the state-of-the-art Mixed-Integer Quadratic Programming (MIQP). The model predictive control (MPC) utilizes the gait generated by the MCTS to optimize the ground reaction forces and future footholds position. The simulation results, performed on a quadruped robot, showed that the proposed framework could generate known periodic gait and adapt the contact sequence to the encountered conditions, including external forces and terrain with unknown and variable properties. When tested on robots with different layouts, the system has also shown its reliability.

This paper introduces a design method for an efficient and agile quadruped robot. A mixed-integer optimization formulation including the number of gear teeth is derived to obtain the optimal gear ratio that minimizes cost for a running-trot with the target speed of 3 m/s. With the inclusion of integer constraints related to the number of gear teeth, detailed design considerations of gear trains can be included in the optimization process. Thermal dissipation of the motor controller is also taken into account in the optimization to consider heat generation during high-speed running. KAIST Hound, a 45 kg robot, designed with the obtained design parameters has successfully demonstrated a 3 m/s running-trot using a nonlinear model predictive controller (NMPC). Furthermore, the robot has proved its robustness by the demonstration of additional experiments such as 22° slope climbing, 3.2 km walking, and traversing a 35 cm obstacle.

This paper presents the design and experimental evaluations of an articulated robotic limb called Capler-Leg. The key element of Capler-Leg is its single-stage cable-pulley transmission combined with a high-gap radius motor. Our cable-pulley system is designed to be as light-weight as possible and to additionally serve as the primary cooling element, thus significantly increasing the power density and efficiency of the overall system. The total weight of active elements on the leg, i.e. the stators and the rotors, contribute more than 60 % of the total leg weight, which is an order of magnitude higher than most existing robots. The resulting robotic leg has low inertia, high torque transparency, low manufacturing cost, no backlash, and a low number of parts. The Capler-Leg system itself, serves as an experimental setup for evaluating the proposed cable-pulley design in terms of robustness and efficiency. A continuous jump experiment shows a remarkable 96.5 % recuperation rate, measured at the battery output. This means that almost all the mechanical energy output during push-off is returned back to the battery during touch-down.

This paper presents a framework which allows a quadrupedal robot to execute dynamic gaits including trot, pace and dynamic lateral walk, as well as a smooth transition between them. Our method relies on an online ZMP based motion planner which continuously updates the reference motion trajectory as a function of the contact schedule and the state of the robot. The planner is coupled with a hierarchical whole-body controller which optimizes the whole-body motion and contact forces by solving a cascade of prioritized tasks. We tested our framework on ANYmal, a fully torque controllable quadrupedal robot which is actuated by series-elastic actuators.

This paper introduces ANYmal, a quadrupedal robot that features outstanding mobility and dynamic motion capability. Thanks to novel, compliant joint modules with integrated electronics, the 30 kg, 0.5 m tall robotic dog is torque controllable and very robust against impulsive loads during running or jumping. The presented machine was designed with a focus on outdoor suitability, simple maintenance, and user-friendly handling to enable future operation in real world scenarios. Performance tests with the joint actuators indicated a torque control bandwidth of more than 70 Hz, high disturbance rejection capability, as well as impact robustness when moving with maximal velocity. It is demonstrated in a series of experiments that ANYmal can execute walking gaits, dynamically trot at moderate speed, and is able to perform special maneuvers to stand up or crawl very steep stairs. Detailed measurements unveil that even full-speed running requires less than 280 W, resulting in an autonomy of more than 2 h.

Legged robots require a robust and fast responding feet contact detection strategy. Common force sensors are often too heavy and can be easily damaged during impacts with the terrain. Therefore, it is desirable to detect a contact without a force sensor. This paper introduces a probabilistic contact detection strategy which considers full dynamics and differential/forward kinematics to maximize the use of available information for contact estimation. This papers shows that such strategy is much more accurate than the state-of-the-art strategy that only take one measure into account, with a quadrupedal robot.

Methods of optimizing a single trajectory are mature enough for planning in many applications. Yet such optimization methods applied to high Degree-Of-Freedom robots either consume too much time to be real-time or approximate the dynamics such that they lack physical consistency. In this paper, we present a method of precomputing optimized trajectories and compressing the information to get a compact representation of the optimal policy function. By varying the initial configuration of a robot and optimizing multiple trajectories, the controller gains knowledge about the optimal policy function. Such computation can be performed on a powerful workstation or even supercomputers instead of an onboard computer of the robot. The precomputed optimal trajectories are stored in a Single-hidden Layer Feedforward neural Network (SLFN) using Optimally Pruned Extreme Learning Machine (OP-ELM). This ensures minimal representation of the model and fast evaluation of the SLFN. We first explain our method using a simple time-optimal control problem with an analytical solution. We then demonstrate how this method can work even for high dimensional state by optimizing a foothold strategy of a full quadruped robot in simulation.

In this paper we present a method for fusing optical flow and inertial measurements. To this end, we derive a novel visual error term which is better suited than the standard continuous epipolar constraint for extracting the information contained in the optical flow measurements. By means of an unscented Kalman filter (UKF), this information is then tightly coupled with inertial measurements in order to estimate the egomotion of the sensor setup. The individual visual landmark positions are not part of the filter state anymore. Thus, the dimensionality of the state space is significantly reduced, allowing for a fast online implementation. A nonlinear observability analysis is provided and supports the proposed method from a theoretical side. The filter is evaluated on real data together with ground truth from a motion capture system.