CNNs based Foothold Selection for Energy-Efficient Quadruped Locomotion over Rough Terrains

Lu, Chen; Ye, Shusheng; Sun, Chuen-Tsai; Zhang, Aidong; Deng, Ganyu; Liao, Tianjiao; Sun, Junwen

doi:10.1109/robio49542.2019.8961842

Cited by 10 publications

(4 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A linear regression method was used to approximate the selection of an expert user to adapt the landing location of the feet within the template and the motion of the base was designed to follow these footholds. Inspired by this, methods have resourced to template-based foothold selection using learning-based [8], [15], [24], [21], [22], [23] or fast optimization [25], [26] strategies. In this way the optimization problem becomes lighter.…”

Section: Related Workmentioning

confidence: 99%

Foothold Evaluation Criterion for Dynamic Transition Feasibility for Quadruped Robots

Clemente¹,

Villarreal²,

Bratta³

et al. 2022

Preprint

View full text Add to dashboard Cite

To traverse complex scenarios reliably a legged robot needs to move its base aided by the ground reaction forces, which can only be generated by the legs that are momentarily in contact with the ground. A proper selection of footholds is crucial for maintaining balance. In this paper, we propose a foothold evaluation criterion that considers the transition feasibility for both linear and angular dynamics to overcome complex scenarios. We devise convex and nonlinear formulations as a direct extension of [1] in a receding-horizon fashion to grant dynamic feasibility for future behaviours. The criterion is integrated with a Vision-based Foothold Adaptation (VFA) strategy that takes into account the robot kinematics, leg collisions and terrain morphology. We verify the validity of the selected footholds and the generated trajectories in simulation and experiments with the 90kg quadruped robot HyQ.

show abstract

Section: Related Workmentioning

confidence: 99%

Foothold Evaluation Criterion for Dynamic Transition Feasibility for Quadruped Robots

Clemente¹,

Villarreal²,

Bratta³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Related work: Bazeille et al [17] (2013) and Siravuru et al [18] (2017) proposed methods of using cameras to predict the safe footholds which includes a large CNN processing cost. [19] used a recurrent policy to encode directly the visual perception from LiDAR sensing and proprioceptive information, and Chen et al [20] proposed a Convolutional Neural Networks based classifier to plan optimal quadrupedal footholds. Recently, Gangapurwala et al [21] (2021) proposed a hybrid method of RL and trajectory optimization algorithm that takes terrain height map to generate an adaptive joint-level trajectories, and a similar DRL approach was proposed by [22].…”

Section: Introductionmentioning

confidence: 99%

Learning Vision-Guided Dynamic Locomotion Over Challenging Terrains

Liu¹,

Acero²,

Li³

2021

Preprint

View full text Add to dashboard Cite

Legged robots are becoming increasingly powerful and popular in recent years for their potential to bring the mobility of autonomous agents to the next level. This work presents a deep reinforcement learning approach that learns a robust Lidar-based perceptual locomotion policy in a partially observable environment using Proximal Policy Optimisation. Visual perception is critical to actively overcome challenging terrains, and to do so, we propose a novel learning strategy: Dynamic Reward Strategy (DRS), which serves as effective heuristics to learn a versatile gait using a neural network architecture without the need to access the history data. Moreover, in a modified version of the OpenAI gym environment, the proposed work is evaluated with scores over 90% success rate in all tested challenging terrains.

show abstract

“…Alternatively, optimization-based algorithms 4,5 allow online body motion and foothold replanning using terrain maps but their computational complexity limits their application to slow (quasi-static) gaits. For this reason, some approaches [6][7][8][9] use machine learning models to approximate the computationally heavy operations related to terrain analysis. A convolutional neural network (CNN), for example, can be used to evaluate the geometry of terrains and identify rough areas to avoid foot and leg collisions.…”

Section: Introductionmentioning

confidence: 99%

“…A convolutional neural network (CNN), for example, can be used to evaluate the geometry of terrains and identify rough areas to avoid foot and leg collisions. [7][8][9] In a previous work, we proposed a CNN-based foothold adaptation strategy to infer safe foothold locations from terrain data in approximately 0.2 ms, which allowed us to apply the approach for dynamic locomotion. 7 However, this approach assumed that the robot moves with a constant velocity during the entire swing phase which has some disadvantages in scenarios similar to Fig.…”

Section: Introductionmentioning

confidence: 99%

On the influence of body velocity in foothold adaptation for dynamic legged locomotion via CNNs

2020

Robots in Human Life

View full text Add to dashboard Cite

This paper analyzes the importance of considering the body velocity for foothold adaptation during legged locomotion. We show how this velocity affects the decisions made by a foothold adaption strategy, and the number of feasible footholds computed by the approach. We extend our previous work by considering the body velocity in this foothold adaptation method and augmenting a convolutional neural network (CNN) classifier to account for the current velocity of the robot. Our results suggest that the foothold evaluation strategy has a better performance with this new CNN than with architectures that assume a constant velocity and that only consider heightmaps as input for the foothold evaluation.

show abstract

CNNs based Foothold Selection for Energy-Efficient Quadruped Locomotion over Rough Terrains

Cited by 10 publications

References 22 publications

Foothold Evaluation Criterion for Dynamic Transition Feasibility for Quadruped Robots

Foothold Evaluation Criterion for Dynamic Transition Feasibility for Quadruped Robots

Learning Vision-Guided Dynamic Locomotion Over Challenging Terrains

On the influence of body velocity in foothold adaptation for dynamic legged locomotion via CNNs

Contact Info

Product

Resources

About