#Robot perception (10)

Waterfall Pagination

0:37

InFlux++: benchmark and synthetic data for dynamic camera intrinsics

@ErichLiang · 208 views · 2026-09-01

InFluxcamera intrinsicsBenchmarks

4:25

DART turns SAM3 into a real-time open-vocabulary detector at 55.8 AP on COCO

@rsasaki0109 · 216 views · 2026-08-29

open-vocabulary detectionSAM3DART

Hesai JT128 Mini Hyper-Hemispherical 360° LiDAR

The flagship of Hesai's JT-series robot LiDARs, featuring 128 independent transceiver channels, up to 1.15M points/s single-return point rate and a 360°x89° hyper-hemispherical FOV; already chosen by 50+ robotics companies including Unitree and Ant Lingbo, making it one of the mainstream 360° LiDARs for robotics at WRC 2026.

COMPONENT

SensorsLiDARHesai JT128360° SuperhemisphericalRobot perception

RoboSense Airy 192-Beam Hemispherical Digital LiDAR

The world's first 192-beam hemispherical digital LiDAR, built on VCSEL emitter chips and a high-sensitivity SPAD-SoC digital detection architecture with chip-based transceivers and lossless echo data. A 360°x90° hemispherical FOV covers a 60 m radius at 1.72M pts/s with 1cm@1σ precision; just φ60xH63 mm, <240 g and <8 W with an integrated IMU. Structured, algorithm-friendly point clouds serve SLAM, obstacle avoidance and navigation for courtyard, quadruped and delivery robots in all weather.

COMPONENT

LiDARDigital LiDARRoboSense AiryRobot perceptionSLAM

See like a Robot: Robot-Centric Pointmaps for Vision-Language-Action Models

This paper addresses the frame mismatch in VLA models between camera-frame observation and robot-frame action by introducing robot-centric pointmaps—images whose pixels store 3D coordinates in the robot frame. Pointmaps provide robot-frame 3D geometry while preserving image structure, enabling cross-viewpoint generalization across diverse camera setups.

Byungkun Lee, Dongyoon Hwang, Dongjin KimJul 13, 2026

VLAPointmapcross-viewpointJul 13, 2026

1:25

Robot-centric pointmaps align VLA perception with the robot frame

@shenzhenfoundry · 54 views · 2026-08-12

VLARobot-Centric PointmapsEmbodied AI

Describe Anything, Anywhere, at Any Moment (DAAAM)

DAAAM is a spatio-temporal memory framework for real-time 4D scene understanding. It uses an optimization-based frontend to infer detailed semantic descriptions from localized captioning models (DAM) with batch processing, and builds a hierarchical 4D scene graph for large-scale AR and robot autonomy applications.

Nicolas Gorlo, Lukas Schmid, Luca CarloneNov 29, 2025

spatio-temporalDAMMIT-SPARKNov 29, 2025

0:30