Learning robust perceptive locomotion for quadrupedal robots in the wild

By Takahiro Miki, Jemin Hwangbo, Lorenz Wellhausen, Vladlen Koltun, Marco Hutter. Science Robotics, 2022

1,117 citations recorded by the source corpus.

Legged robots that can operate autonomously in remote and hazardous environments will greatly increase opportunities for exploration into underexplored areas. Exteroceptive perception is crucial for fast and energy-efficient locomotion: Perceiving the terrain before making contact with it enables planning and adaptation of the gait ahead of time to maintain speed and stability. However, using exteroceptive perception robustly for locomotion has remained a grand challenge in robotics. Snow, vegetation, and water visually appear as obstacles on which the robot cannot step or are missing altogether due to high reflectance. In addition, depth perception can degrade due to difficult lighting, dust, fog, reflective or transparent surfaces, sensor occlusion, and more. For this reason, the most robust and general solutions to legged locomotion to date rely solely on proprioception. This severely limits locomotion speed because the robot has to physically feel out the terrain before adapting its gait accordingly. Here, we present a robust and general solution to integrating exteroceptive and proprioceptive perception for legged locomotion. We leverage an attention-based recurrent encoder that integrates proprioceptive and exteroceptive input. The encoder is trained end to end and learns to seamlessly combine the different perception modalities without resorting to heuristics. The result is a legged locomotion controller with high robustness and speed. The controller was tested in a variety of challenging natural and urban environments over multiple seasons and completed an hour-long hike in the Alps in the time recommended for human hikers. Description A legged locomotion controller achieves high robustness and speed in the wild by combining multimodal information.

Subfields: Aerospace Engineering, Biomedical Engineering, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/5da5c2167a85ecb5d1ea22656ae36fdf995df0f2

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Aerospace Engineering

In Biomedical Engineering

In Computer Vision and Pattern Recognition

Citations in the corpus

Robot Parkour Learning (2023, background): “…have shown robust performance on walking [8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 12, 21, 22, 23, 24, 25, 26, 27], climbing stairs [20, 28, 29, 30, 31, 32, 33], mimicking animals [34, 35, 36, 37, 38, 39] and legged mobile manipulation [40, 41, 42] by learning a policy in simulation…”
HumanPlus: Humanoid Shadowing and Imitation from Humans (2024, methodology): “To tackle the control complexity of humanoids, we follow the recent success in legged robotics using large-scale reinforcement learning in simulation and sim-to-real transfer [41, 51] to train a low-level policy for whole-body control.”
ANYmal parkour: Learning agile navigation for quadrupedal robots (2023, methodology): “…must traverse various irregular terrains consisting of stairs, slopes, and randomly placed small obstacles, similar to the ones commonly used in previous legged locomotion works [7], [41]. b) Jumping: The robot starts on a box and must jump to a neighboring box separated by a gap of up to 1 m .”
Reaching the limit in autonomous racing: Optimal control versus reinforcement learning (2023, background): “RL has arisen as an attractive alternative to conventional controller design, demonstrating exceptional performance in various domains, such as quadrupedal locomotion over challenging terrain (26, 27).”
Orbit: A Unified Simulation Framework for Interactive Robot Learning Environments (2023, background): “Methods such as reinforcement learning (RL) have shown incredible success in challenging problems such as quadrupedal locomotion [1], [2], [3] and”
Legged Locomotion in Challenging Terrains using Egocentric Vision (2022, methodology): “Competing approaches which rely on metric localization to construct 44 elevation maps either assume access to perfect elevation map in the real world, or learn a system 45 capable of handling noise which then is forced to rely primarily on proprioception while using vision 46 for only some additiona”
BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Dataset (2023, methodology): “Previously, the PPO algorithm has been effectively applied across a diverse range of domains, such as computer vision [50, 51, 52] and robotics [53, 54, 55].”
Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes (2024, methodology): “Exteroceptive observations are typically in the form of terrain height maps (35, 36), depth images (44, 45), and RGB images (43).”
Extreme Parkour with Legged Robots (2023, neutral): “Due to the noisy and laggy action, the perception has artifacts, latency, and jitter [9].”
DayDreamer: World Models for Physical Robot Learning (2022, background): “…Lee et al., 2020; Irpan et al., 2020; Rudin et al., 2021; Kumar et al., 2021; Siekmann et al., 2021; Smith et al., 2021; Escontrela et al., 2022; Miki et al., 2022), leverage 0 40 80 120 Minutes 0.1 0.3 0.5 0.7 0.9 A v g D i s t t o G o a l Figure 7: Sphero Navigation This task requires the…”