Inquiry about LIKO Method Applicability for Quadruped Robots Climbing Stairs

[zma69650]

Hi,

I have recently reviewed your paper "LIKO: LiDAR, Inertial, and Kinematic Odometry for Bipedal Robots" and found it very insightful. Thank you for sharing your work and making it open-source.

I am currently working on a project involving quadruped robots, and I am particularly interested in whether the LIKO method could be applied to enhance their ability to climb stairs. Additionally, I would like to understand how well this method estimates the height of the robot, especially in dynamic environments like stair climbing.

Could you please provide some insights into:

Whether the LIKO method has the potential to be adapted for quadruped robots? How effective is the height estimation in such scenarios? Any modifications or considerations that might be necessary for implementing LIKO in quadruped robots? Thank you very much for your time and assistance. I look forward to your response.

Best regards

[Mr-Zqr]

Hi,

Thanks very much for your interest in our work. We also got a suggestion about adding some stair scenarios to our dataset from reviewers. Unfortunately, our robot did not have the ability to climb consecutive stairs so we ended up not including such a challenging scenario. Nonetheless, we tested the LIKO with the robot walking up and down a slope, and the algorithm performed quite well.

Here are my thoughts on your questions:

1. Yes, but some changes are needed. Such as functions UpdateKinematicState, UpdateContactPositionAftSwitch, and KinematicVelocityEstimation to support the position and velocity of four legs, as well as the contact detection part.
2. The algorithm performed well for slope in a room, so I believe it will also be robust for stairs in open areas. I am not sure how the algorithm will perform in a narrower scenario such as spiral staircases or those consecutive stairs in the apartments. But as long as there are enough planes for the LiDAR to perform registration, it should be fine.
3. Apart from the modifications mentioned in "1.", I think it is also better to realize a dynamic number of contact points to better suit the nature of quadruped robots.

Thanks for reaching out, please let me know if you have any further questions.

Best regards, Qingrui

[zma69650]

感谢回复！！我在UpdateContactPositionAftSwitch函数中，我更新state_point.pc中的foot，用的是触地腿位置的平均值， UpdateBodyPositionKine函数也是，不知道对不对，还有请问一下R_base_foot函数怎么更新呢，感谢回复！！

[Mr-Zqr]

抱歉回复晚了！

您是指算出同一时刻下所有触地腿接触点位置的平均值，当作此时的接触点进行更新吗？感觉应该可以，只是这样后续触地腿不变时的正运动学更新也需要计算平均值，在平地上应该没啥问题，崎岖地面或者楼梯环境不知道效果如何，您可以测试一下。

R_base_foot这里是一个失误，在仿人上应当是根据关节码盘的数据使用pinocchio进行计算的。这个量在算法中是预测阶段矩阵F_w中的一项，用于预测被建模为随机游走的接触点的位置。对于点足，初始化为单位矩阵即可。