SummaryThis article addresses the problem of maneuvering multiple agents that must visit a number of target sets, while enforcing connectivity constraints and avoiding obstacle as well as interagent collisions. The tool to cope with the problem is a formulation of model predictive control including binary decision variables. In this regard, two mixed‐integer linear programming formulations are presented, considering a trade‐off between optimality and scalability between them. Simulation results are also shown to illustrate the main features of the proposed approaches.

No abstract

No abstract

Summary This paper presents a mixed‐integer model predictive controller for walking. In the proposed scheme, mixed‐integer quadratic programs (MIQP) are solved online to simultaneously decide center of mass jerks, footsteps positions, durations, and rotations while respecting actuation, geometry, and contact constraints. Most walking controllers require preplanned footstep rotations to avoid dealing with the nonlinearity introduced by foot rotation decision. The main contribution of this work is an optimization formulation where feet rotations are automatically planned to attain a reference speed rotation. Finally, simulation results are shown to present and discuss the capabilities of the proposed formulation.

With the widespread adoption of deep learning, reinforcement learning (RL) has experienced a dramatic increase in popularity, scaling to previously intractable problems, such as playing complex games from pixel observations, sustaining conversations with humans, and controlling robotic agents. However, there is still a wide range of domains inaccessible to RL due to the high cost and danger of interacting with the environment. Offline RL is a paradigm that learns exclusively from static datasets of previously collected interactions, making it feasible to extract policies from large and diverse training datasets. Effective offline RL algorithms have a much wider range of applications than online RL, being particularly appealing for real-world applications, such as education, healthcare, and robotics. In this work, we contribute with a unifying taxonomy to classify offline RL methods. Furthermore, we provide a comprehensive review of the latest algorithmic breakthroughs in the field using a unified notation as well as a review of existing benchmarks' properties and shortcomings. Additionally, we provide a figure that summarizes the performance of each method and class of methods on different dataset properties, equipping researchers with the tools to decide which type of algorithm is best suited for the problem at hand and identify which classes of algorithms look the most promising. Finally, we provide our perspective on open problems and propose future research directions for this rapidly growing field.

No abstract

Controlling a humanoid robot with its typical many degrees of freedom is a complex task, and many methods have been proposed to solve the problem of humanoid locomotion. In this work, we generate a gait for a Hitec Robonova-I robot using a model-free approach, where fairly simple parameterized models, based on truncated Fourier series, are applied to generate joint angular trajectories. To find a parameter set that generates a fast and stable walk, optimization algorithms were used, specifically a genetic algorithm and particle swarm optimization. The optimization process was done in simulation first, and the learned walk was then adapted to the real robot. The simulated model of the Robonova-I was made using the USARSim simulator, and tests made to evaluate the resulting walks verified that the best walk obtained is faster than the ones publicly available for the Robonova-I. Later, to provide an additional validation, the same process was carried out for the simulated Nao from the RoboCup 3D Soccer Simulation League. Again, the resulting walk was fast and stable, overcoming the speed of the publicly available magma-AF base team.

No abstract