LoRA: Low-Rank Adaptation of Large Language Models
Low-rank trainable updates adapt a frozen pretrained model without updating all of its parameters.
THE RESEARCH ATLAS
PointWalker: Mapless Quadruped Locomotion Directly From Raw LiDAR Point Clouds
Fuses LiDAR point clouds and proprioception through attention, connecting terrain perception with a reinforcement-learned quadruped policy.
Institution and location sources
Location refers to credited institutions or the listed organization, not author citizenship or an exhaustive collaboration list.
Why read this
Study physical intelligence without a language-dialogue interface: perception, body state and contact-rich movement must form a real-time loop.
What it connects
A PPO-trained policy connects point-cloud attention and proprioception on the X30. Author affiliations include Zhejiang University and Deep Robotics; this is joint research, not a solely corporate paper.
Preprint; simulation and robot experiments
The preprint reports simulation and hardware tests. Transfer beyond the sensor layout, global navigation and moving supports remain limited or open.
Follow these connections ↗Sources for relations and versions
Citation
@misc{li2026pointwalker,
title = {PointWalker: Mapless Quadruped Locomotion Directly From Raw LiDAR Point Clouds},
author = {Sicen Li and Zhen Chu and Chao Li and Jun Zheng and Qiuguo Zhu and Jun Wu},
year = {2026},
eprint = {2610.15548},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.15548}
}Rights remain with the original authors. Refer to the original source for citations and licensing.
Related research
Training Compute-Optimal Large Language Models
The study examines how to allocate a training-compute budget between model size and training data.
FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness
An IO-aware attention algorithm reduces memory transfers while computing exact attention.