MPC-based Controller with Terrain Insight for Dynamic Legged Locomotion

Villarreal, Octavio; Barasuol, Victor; Wensing, Patrick M.; Caldwell, Darwin G.; Semini, Claudio

doi:10.48550/arxiv.1909.13842

Cited by 4 publications

(6 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In recent years, machine learning has played an increasingly important role in quadruped robot control. Villarreal et al [7] combined a convolutional neural network with MPC. Yang et al [8] used a gating neural network combined with pretrained neural networks to adapt to changes in the environment and tested the combined method on a real robot.…”

Section: Related Workmentioning

confidence: 99%

Imitation and Adaptation Based on Consistency: A Quadruped Robot Imitates Animals from Videos Using Deep Reinforcement Learning

Yao¹,

Wang²,

Yang³

et al. 2022

Preprint

View full text Add to dashboard Cite

The essence of quadrupeds' movements is the movement of the center of gravity, which has a pattern in the action of quadrupeds. However, the gait motion planning of the quadruped robot is time-consuming. Animals in nature can provide a large amount of gait information for robots to learn and imitate. Common methods learn animal posture with a motion capture system or numerous motion data points. In this paper, we propose a video imitation adaptation network (VIAN) that can imitate the action of animals and adapt it to the robot from a few seconds of video. The deep learning model extracts key points during animal motion from videos. The VIAN eliminates noise and extracts key information of motion with a motion adaptor, and then applies the extracted movements function as the motion pattern into deep reinforcement learning (DRL). To ensure similarity between the learning result and the animal motion in the video, we introduce rewards that are based on the consistency of the motion. DRL explores and learns to maintain balance from movement patterns from videos, imitates the action of animals, and eventually, allows the model to learn the gait or skills from short motion videos of different animals and to transfer the motion pattern to the real robot.

show abstract

Section: Related Workmentioning

confidence: 99%

Imitation and Adaptation Based on Consistency: A Quadruped Robot Imitates Animals from Videos Using Deep Reinforcement Learning

Yao¹,

Wang²,

Yang³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…This task aims to solve real benchmarking problems, train the robot in a simulated environment, and track the robot's movements. 3) Octavio Villarreal, Victor Barasuol et al [3] Proposed a new control strategy for dynamic leg movement. This strategy is based on two main elements.…”

Section: IImentioning

confidence: 99%

“…While these results are impressive, model-free methods have not yet been successfully applied to control physical systems other than research demos. Unfortunately, the current trends in RL research conflict these obstacles [3]. Searching for sample-efficient methods (that is, methods that require little data) generally requires the development of increasingly complex methods.…”

Section: Introductionmentioning

confidence: 99%

Tracking Locomotion using Reinforcement Learning

Jadhav¹

2022

IJRASET

View full text Add to dashboard Cite

This article presents the concept of reinforcement learning, which prepares a static direct approach for consistent control problems, and adjusts cutting-edge techniques for testing effectiveness in benchmark Mujoco locomotion tasks. This model was designed and developed to use the Mujoco Engine to track the movement of robotic structures and eliminate problems with assessment calculations using perceptron’s and random search algorithms. Here, the machine learning model is trained to make a series of decisions. The humanoid model is considered to be one of the most difficult and ongoing problems to solve by applying state-of-the-art RL technology. The field of machine learning has a great influence on the training model of the RL environment. Here we use random seed values to provide continuous input to achieve optimized results. The goal of this project is to use the Mujoco engine in a specific context to automatically determine the ideal behavior of the robot in an augmented reality environment. Enhanced random search was introduced to train linear guidelines for achieving the efficiency of Mujoco roaming tasks. The results of these models highlight the variability of the Mujoco benchmark task and lead to efficiently optimized rewards

show abstract

“…Another common control framework accounts for the full rigid-body-dynamics of the robot (Winkler et al, 2015;Bellicoso et al, 2016), which tends to better capture the robot dynamics, but is more computationally expensive. Lastly, two-stage control frameworks have also been presented with low-rate Model Predictive Control and a high-rate whole-body controller (Villarreal et al, 2019), extending the robustness of a simple whole-body control framework by accounting for a longer time horizon.…”

Section: Introductionmentioning

confidence: 99%

Quadrupedal Walking over Complex Terrain with a Quasi-Direct Drive Actuated Robot

Griffin¹,

McCrory²,

Bertrand³

et al. 2022

View full text Add to dashboard Cite

In this paper, we present our approach to achieve autonomous walking over complex terrain on the quadrupedal robot, LLAMA. LLAMA is a prototype robot designed by NASA Jet Propulsion Lab as part of the Army Research Laboratory’s Robotics Collaborative Technology Alliance. One of the major objectives of this robot is to be capable of traversing complex terrain autonomously to enable operating alongside a human squadron. This goal requires the robot to be capable of identifying practical footholds according to the environment, which may be sparse, and using these for walking. To accomplish this end, we first introduce two new contact planners. One is based on either desired body path plans; the other an A* graph-search based planner that find contacts over rough terrain given the environmental constraints. We then plan a dynamic trajectory using a custom Divergent Component of Motion planner, which is tracked using a whole-body inverse-dynamics control framework. We also introduce new methods for maintaining balance by adjusting step position and timing. We additionally discuss our approach for contact detection without the use of force sensors. We highlight the results in several experiments, both in our laboratory and at the RCTA Capstone Event at the Camp Lejeune Marine Corps base. We conclude with a discussion of these results, specific implementation problems, and lessons learned when developing such a control architecture on a quadruped.

show abstract

MPC-based Controller with Terrain Insight for Dynamic Legged Locomotion

Cited by 4 publications

References 0 publications

Imitation and Adaptation Based on Consistency: A Quadruped Robot Imitates Animals from Videos Using Deep Reinforcement Learning

Imitation and Adaptation Based on Consistency: A Quadruped Robot Imitates Animals from Videos Using Deep Reinforcement Learning

Tracking Locomotion using Reinforcement Learning

Quadrupedal Walking over Complex Terrain with a Quasi-Direct Drive Actuated Robot

Contact Info

Product

Resources

About