High-frequency control in continuous action and state spaces is essential for practical applications in the physical world. Directly applying end-to-end reinforcement learning to high-frequency control tasks struggles with assigning credit to actions across long temporal horizons, compounded by the difficulty of efficient exploration. The alternative, learning low-frequency policies that guide higher-frequency controllers (e.g., proportional-derivative (PD) controllers), can result in a limited total expressiveness of the combined control system, hindering overall performance. We introduce EvoControl, a novel bi-level policy learning framework for learning both a slow high-level policy (using PPO) and a fast low-level policy (using Evolution Strategies) for solving continuous control tasks. Learning with Evolution Strategies for the lower-policy allows robust learning for long horizons that crucially arise when operating at higher frequencies. This enables EvoControl to learn to control interactions at a high frequency, benefitting from more efficient exploration and credit assignment than direct high-frequency torque control without the need to hand-tune PD parameters. We empirically demonstrate that EvoControl can achieve a higher evaluation reward for continuous-control tasks compared to existing approaches, specifically excelling in tasks where high-frequency control is needed, such as those requiring safety-critical fast reactions.

We demonstrate the ability of large language models (LLMs) to perform iterative self-improvement of robot policies. An important insight of this paper is that LLMs have a built-in ability to perform (stochastic) numerical optimization and that this property can be leveraged for explainable robot policy search. Based on this insight, we introduce the SAS Prompt (Summarize, Analyze, Synthesize) – a single prompt that enables iterative learning and adaptation of robot behavior by combining the LLM's ability to retrieve, reason and optimize over previous robot traces in order to synthesize new, unseen behavior. Our approach can be regarded as an early example of a new family of explainable policy search methods that are entirely implemented within an LLM. We evaluate our approach both in simulation and on a real-robot table tennis task. Project website: sites.google.com/asu.edu/sas-llm/

Achieving human-level performance on real world tasks is a north star for the robotics community. We present the first learned robot agent that reaches amateur humanlevel performance in competitive table tennis. Table tennis is a physically demanding sport that takes humans years to master. We contribute (1) a hierarchical and modular policy architecture consisting of (i) low level controllers with their skill descriptors that model their capabilities and (ii) a high level controller that chooses the low level skills, (2) techniques for enabling zero-shot sim-to-real and curriculum building, including an iterative approach (train in sim, deploy in real), and (3) real time adaptation to unseen opponents. Policy performance was assessed through 29 robot vs. human matches of which the robot won 45 % (13/29). All humans were unseen players and their skill level varied from beginner to tournament level. Whilst the robot lost all matches vs. the most advanced players it won 100 % matches vs. beginners and 55 % matches vs. intermediate players, demonstrating solidly amateur humanlevel performance. Videos of the matches can be viewed here1.See sites https://google.com/view/competitive-robot-table-tennis.

In this paper, we describe the design of the Barkour benchmark for measuring robot agility in navigating complex environments. Despite the growing interest in developing agile robot locomotion skills, the field lacks systematic benchmarks to measure the performance of robotic control systems and hardware in agility-focused tasks. This motivated us to propose the Barkour benchmark, an obstacle course designed to quantify agility across various robotic platforms. Inspired by dog agility competitions, the course features diverse obstacles and a time-based scoring mechanism, encouraging researchers to develop controllers that enable robots to move quickly, precisely, and with adaptability. This benchmark is challenging as it demands diverse motion skills and the time-based scoring requires control precision at high speed. Along with the design details presented in the paper, we release our simulated environment setups in MuJoCo-XLA and the CAD model of a custom-designed quadruped robot to facilitate future research to reproduce the Barkour setup (available at sites.google.com/view/barkour). We hope these together will accelerate the pace of robot agility research.

Training a high-dimensional simulated agent with an under-specified reward function often leads the agent to learn physically infeasible strategies that are ineffective when deployed in the real world. To mitigate these unnatural behaviors, reinforcement learning practitioners often utilize complex reward functions that encourage physically plausible behaviors. However, a tedious labor-intensive tuning process is often required to create hand-designed rewards which might not easily generalize across platforms and tasks. We propose substituting complex reward functions with “style rewards” learned from a dataset of motion capture demonstrations. A learned style reward can be combined with an arbitrary task reward to train policies that perform tasks using naturalistic strategies. These natural strategies can also facilitate transfer to the real world. We build upon Adversarial Motion Priors - an approach from the computer graphics domain that encodes a style reward from a dataset of reference motions - to demonstrate that an adversarial approach to training policies can produce behaviors that transfer to a real quadrupedal robot without requiring complex reward functions. We also demonstrate that an effective style reward can be learned from a few seconds of motion capture data gathered from a German Shepherd and leads to energy-efficient locomotion strategies with natural gait transitions.

Deep Q learning (DQN) has enabled robot agents to accomplish vision based tasks that seemed out of reach. Despite recent success stories, there are still several sources of computational complexity that challenge the performance of DQN. We place the focus on vision manipulation tasks, where the correct action selection is often predicated on a small number of pixels. We observe that in some of these tasks DQN does not converge to the optimal Q function, and their values do not separate well optimal and suboptimal actions. In consequence, the policies obtained with DQN tend to be brittle and manifest a low success rate, especially in long horizon tasks. In this work we show the benefits of Reward Machines (RMs) for Deep Q learning (DQRM) in vision based robot manipulation tasks. Reward machines decompose the task at an abstract level, inform the agent about their current stage along task completion, and guide them via dense rewards. We show that RMs help DQN learn the optimal Q values in each abstract state. Their policies are more robust, manifest higher success rate, and are learned with fewer training steps compared with DQN. The benefits of RMs are more evident in long-horizon tasks, where we show that DQRM is able to learn good-quality policies with six times times fewer training steps than DQN, even when this is equipped with dense reward shaping.

Developing agile behaviors for legged robots re-mains a challenging problem. While deep reinforcement learning is a promising approach, learning truly agile behaviors typically requires tedious reward shaping and careful curriculum design. We formulate agile locomotion as a multi-stage learning problem in which a mentor guides the agent throughout the training. The mentor is optimized to place a checkpoint to guide the movement of the robot’s center of mass while the student (i.e. the robot) learns to reach these checkpoints. Once the student can solve the task, we teach the student to perform the task without the mentor. We evaluate our proposed learning system with a simulated quadruped robot on a course consisting of randomly generated gaps and hurdles. Our method significantly outperforms a single-stage RL baseline without a mentor, and the quadruped robot can agilely run and jump across gaps and obstacles. Finally, we present a detailed analysis of the learned behaviors’ feasibility and efficiency.

Legged locomotion is a challenging task for learning algorithms, especially when the task requires a diverse set of primitive behaviors. To solve these problems, we introduce a hierarchical framework that can automatically learn to decompose complex locomotion tasks. A high-level policy issues commands in the form of a latent vector and also selects for how long the low-level policy will execute the latent command. Concurrently, the low-level policy uses the latent command and only the robot's on-board sensors to control the robot's actuators. Our approach allows the high-level policy to run at a lower frequency than the low-level one. We test our framework on a path-following task for a dynamic quadruped robot and we show that steering behaviors automatically emerge in the latent command space as low-level skills are needed for this task. We then show efficient adaptation of the trained policy to new tasks by transfer of the trained low-level policy. Finally, we validate the policies on a real quadruped robot. To the best of our knowledge, this is the first application of end-to-end hierarchical learning to a real robotic locomotion task.

We propose a simple drop-in noise-tolerant replacement for the standard finite difference procedure used ubiquitously in blackbox optimization. In our approach, parameter perturbation directions are defined by a family of structured orthogonal matrices. We show that at the small cost of computing a Fast Walsh-Hadamard/Fourier Transform (FWHT/FFT), such structured finite differences consistently give higher quality approximation of gradients and Jacobians in comparison to vanilla approaches that use coordinate directions or random Gaussian perturbations. We find that trajectory optimizers like Iterative LQR and Differential Dynamic Programming require fewer iterations to solve several classic continuous control tasks when our methods are used to linearize noisy, blackbox dynamics instead of standard finite differences. By embedding structured exploration in a quasi-Newton optimizer (LBFGS), we are able to learn agile walking and turning policies for quadruped locomotion, that successfully transfer from simulation to actual hardware. We theoretically justify our methods via bounds on the quality of gradient reconstruction and provide a basis for applying them also to nonsmooth problems.

Soft robots offer many advantages over traditional rigid robots. However, soft robots can be difficult to control with standard control methods. Fortunately, evolutionary algorithms can offer an elegant solution to this problem. Instead of creating controls to handle the intricate dynamics of these robots, we can simply evolve the controls using a simulation to provide an evaluation function. In this article, we show how such a control paradigm can be applied to an emerging field within soft robotics: robots based on tensegrity structures. We take the model of the Spherical Underactuated Planetary Exploration Robot ball (SUPERball), an icosahedron tensegrity robot under production at NASA Ames Research Center, develop a rolling locomotion algorithm, and study the learned behavior using an accurate model of the SUPERball simulated in the NASA Tensegrity Robotics Toolkit. We first present the historical-average fitness-shaping algorithm for coevolutionary algorithms to speed up learning while favoring robustness over optimality. Second, we use a distributed control approach by coevolving open-loop control signals for each controller. Being simple and distributed, open-loop controllers can be readily implemented on SUPERball hardware without the need for sensor information or precise coordination. We analyze signals of different complexities and frequencies. Among the learned policies, we take one of the best and use it to analyze different aspects of the rolling gait, such as lengths, tensions, and energy consumption. We also discuss the correlation between the signals controlling different parts of the tensegrity robot.

NASA Ames Research Center is developing a compliant modular tensegrity robotic platform for planetary exploration. In this paper we present the design and evolution of the platform's main hardware component, an untethered, robust tensegrity strut, with rich sensor feedback and cable actuation. Each strut is a complete robot, and multiple struts can be combined together to form a wide range of complex tensegrity robots. Our current goal for the tensegrity robotic platform is the development of SUPERball, a 6-strut icosahedron underactuated tensegrity robot aimed at dynamic locomotion for planetary exploration rovers and landers, but the aim is for the modular strut to enable a wide range of tensegrity morphologies. SUPERball is a second generation prototype, evolving from the tensegrity robot ReCTeR, which is also a modular, lightweight, highly compliant 6-strut tensegrity robot that was used to validate our physics based NASA Tensegrity Robot Toolkit (NTRT) simulator. Many hardware design parameters of the SUPERball were driven by locomotion results obtained in our validated simulator. These evolutionary explorations helped constrain motor torque and speed parameters, along with strut and string stress. As construction of the hardware has finalized, we have also used the same evolutionary framework to evolve controllers that respect the built hardware parameters.

Tensegrity robots are composed of compression elements (rods) that are connected via a network of tension elements (cables). Tensegrity robots provide many advantages over standard robots, such as compliance, robustness, and flexibility. Moreover, sphere-shaped tensegrity robots can provide non-traditional modes of locomotion, such as rolling. While they have advantageous physical properties, tensegrity robots are hard to control because of their nonlinear dynamics and oscillatory nature. In this paper, we present a robust, distributed, and directional rolling algorithm, “flop and roll”. The algorithm uses coevolution and exploits the distributed nature and symmetry of the tensegrity structure. We validate this algorithm using the NASA Tensegrity Robotics Toolkit (NTRT) simulator, as well as the highly accurate model of the physical SUPERBall being developped under the NASA Innovative and Advanced Concepts (NIAC) program. Flop and roll improves upon previous approaches in that it provides rolling to a desired location. It is also robust to both unexpected external forces and partial hardware failures. Additionally, it handles variable terrain (hills up to 33% grade). Finally, results are compatible with the hardware since the algorithm relies on realistic sensing and actuation capabilities of the SUPERBall.

Tensegrity structures (built from interconnected rods and cables) have the potential to offer a revolutionary new robotic design that is light-weight, energy-efficient, robust to failures, capable of unique modes of locomotion, impact tolerant, and compliant (reducing damage between the robot and its environment). Unfortunately robots built from tensegrity structures are difficult to control with traditional methods due to their oscillatory nature, nonlinear coupling between components and overall complexity. Fortunately this formidable control challenge can be overcome through the use of evolutionary algorithms. In this paper we show that evolutionary algorithms can be used to efficiently control a ball shaped tensegrity robot. Experimental results performed with a variety of evolutionary algorithms in a detailed soft-body physics simulator show that a centralized evolutionary algorithm performs 400% better than a hand-coded solution, while the multiagent evolution performs 800% better. In addition, evolution is able to discover diverse control solutions (both crawling and rolling) that are robust against structural failures and can be adapted to a wide range of energy and actuation constraints. These successful controls will form the basis for building high-performance tensegrity robots in the near future.