Quadruped Robot
A bioinspired robot with four legs that replicates the locomotion of quadrupedal animals to navigate complex terrain. Representative examples include Boston Dynamics' Spot and MIT Mini Cheetah. Quadruped robots represent a significant research direction in robotics, particularly in areas of legged locomotion and dynamic balance.
Timeline (last 90 days)
The robot can now climb stairs more reliably and can cross obstacles up to 11 cm high, with a ground clearance of 18 cm, while handling soft ground with reasonable confidence.
Building a quadruped robot with cheap servos keeps costs controllable and is easy to get started with, but dynamic performance and load capacity are inevitably limited.
Each leg of a quadruped robot typically has 2-3 degrees of freedom, and 12 servos is a common quadruped configuration (3 per leg).
In reinforcement-learning gait control for quadruped robots, the policy network typically takes joint positions, joint velocities, body orientation, and foot contact states as observation inputs.
Traditional quadruped robots mostly adopt the trot gait or the bound gait as their basic dynamic gaits.
The Cubic Doggo project introduced random horizontal pushing forces during training, applying random forces of up to 5N to the 3 kg robot to simulate lateral impacts in real environments.
The total delay of a quadruped robot's control chain (from IMU sampling, state estimation, gait planning, inverse kinematics solving, to motor driver execution) is typically on the order of a few milliseconds to a few tens of milliseconds
A quadruped robot performing posture correction on gravel ground using only an IMU may cause a negative feedback loop: the corrective action itself generates new disturbances on the loose ground, manifesting as more wobbling the more it corrects
Increasing computation speed or sampling rate does not always solve the balance problem of quadruped robots on gravel terrain; the real bottleneck is often the physical inertia in the mechanical execution stage
time_horizon cannot be shortened indefinitely; for tasks such as quadruped walking that require multi-step coordinated actions, too short a horizon makes it difficult for the agent to observe a complete gait cycle, and the advantage estimate will be biased due to frequent bootstrapping.
5 more timeline events
All Facts (16)
The robot can now climb stairs more reliably and can cross obstacles up to 11 cm high, with a ground clearance of 18 cm, while handling soft ground with reasonable confidence.
50%UnverifiedBuilding a quadruped robot with cheap servos keeps costs controllable and is easy to get started with, but dynamic performance and load capacity are inevitably limited.
50%UnverifiedEach leg of a quadruped robot typically has 2-3 degrees of freedom, and 12 servos is a common quadruped configuration (3 per leg).
50%UnverifiedIn reinforcement-learning gait control for quadruped robots, the policy network typically takes joint positions, joint velocities, body orientation, and foot contact states as observation inputs.
50%UnverifiedTraditional quadruped robots mostly adopt the trot gait or the bound gait as their basic dynamic gaits.
50%UnverifiedThe Cubic Doggo project introduced random horizontal pushing forces during training, applying random forces of up to 5N to the 3 kg robot to simulate lateral impacts in real environments.
50%UnverifiedThe total delay of a quadruped robot's control chain (from IMU sampling, state estimation, gait planning, inverse kinematics solving, to motor driver execution) is typically on the order of a few milliseconds to a few tens of milliseconds
50%UnverifiedA quadruped robot performing posture correction on gravel ground using only an IMU may cause a negative feedback loop: the corrective action itself generates new disturbances on the loose ground, manifesting as more wobbling the more it corrects
50%UnverifiedIncreasing computation speed or sampling rate does not always solve the balance problem of quadruped robots on gravel terrain; the real bottleneck is often the physical inertia in the mechanical execution stage
50%Unverifiedtime_horizon cannot be shortened indefinitely; for tasks such as quadruped walking that require multi-step coordinated actions, too short a horizon makes it difficult for the agent to observe a complete gait cycle, and the advantage estimate will be biased due to frequent bootstrapping.
50%UnverifiedThe project's ultimate goal is for the complete machine to have 12 degrees of freedom (3 degrees of freedom per leg across 4 legs), with the total weight kept at approximately 3 kg.
50%UnverifiedThe actuator system uses a total gear reduction ratio of approximately 14:1.
50%UnverifiedThe leg currently being tested in the project uses a 2-degree-of-freedom (2 DOF) structure.
50%Unverified四足机器人载人方案以仿生腿替代轮子，兼具主动越障能力与相对自然的运动姿态，且不对使用者身体施加直接机械约束
50%Unverified四足机器人的自我恢复能力来自强化学习的sim-to-real训练范式，通过在仿真环境中摔倒数百万次学习平衡策略
50%UnverifiedA developer open-sourced a project that trains a quadruped robot to imitate the movement of a real dog using only a segment of ordinary mobile phone video.
50%