RoboSignal Daily · 2026-11-04 · Wednesday
ROBOSIGNAL
ADAPT: An End-to-End Framework for Interactive, Text-Conditioned Humanoid Control
The Robotic Systems Lab presents ADAPT, an end-to-end framework for interactive, text-conditioned humanoid control, at the CoRL conference. Unlike dominant pipelines that generate kinematic motions for a separate RL tracker, ADAPT solves humanoid language control end-to-end in a closed-loop manner, allowing the robot to continuously respond to changing commands. The project page is available at https://wuyan01.github.io/ADAPT-project/.
Robot Models
ADAPT: An End-to-End Framework for Interactive, Text-Conditioned Humanoid Control
The Robotic Systems Lab presents ADAPT, an end-to-end framework for interactive, text-conditioned humanoid control, at the CoRL conference. Unlike dominant pipelines that generate kinematic motions for a separate RL tracker, ADAPT solves humanoid language control end-to-end in a closed-loop manner, allowing the robot to continuously respond to changing commands. The project page is available at https://wuyan01.github.io/ADAPT-project/.
lerobot/Cosmos3-Edge-Policy-LIBERO_10
This repository provides the EMA weights for a Cosmos3-Edge policy fine-tuned on the LIBERO-10 benchmark. The model predicts action chunks and future video frames from third-person and wrist camera inputs. It achieves a 95.2% success rate on the LIBERO-10 suite, evaluated with MuJoCo 2.3.7 on an NVIDIA A800-SXM4-80GB GPU cluster. The release includes training configurations, evaluation commands, and licensing information for redistribution.
G0.5 from Galaxea is now available in LeRobot
LeRobot has released the G0.5 vision-language-action model from Galaxea. This 2B parameter model, built on Qwen3.5, predicts subtasks, object bounding boxes, and actions before each action chunk. The integration also enables the annotation pipeline to generate plans, memory, and visual Q&A for datasets, facilitating longer, more complex multi-step setups.
lerobot/Cosmos3-Edge-Policy-LIBERO
This release provides EMA weights for a Cosmos3-Edge policy fine-tuned on four LIBERO suites (Spatial, Object, Goal, and LIBERO-10). The model predicts 16 actions and future video frames from third-person and wrist camera images. Weights correspond to optimizer update 4000 of the training run. Evaluation was conducted with MuJoCo 2.3.7 on NVIDIA A800 GPUs; results are not comparable with MuJoCo 3.x due to contact physics changes. The repository includes training commands, evaluation scripts, and license notices for redistribution.
Research
Tsinghua and NUS Propose ParallelWorld to Boost Embodied Agent Success Rates via Test-Time Pre-verification
Researchers from Tsinghua University and the National University of Singapore introduced ParallelWorld, a method that integrates multi-step future trajectory pre-verification into high-level planning for embodied agents. By exploring candidate action paths and comparing predicted observations with task progress before execution, the system improves success rates on specific benchmarks while acknowledging limitations in handling precise geometric reasoning and physical contact dynamics. The approach uses an adaptive search strategy to balance computational cost with decision uncertainty, allowing agents to correct plans based on real-world feedback after each executed step.
Data & Training
Qingyan Technology Joins Consortium to Release ReAct-Manip Benchmark for Robot 'Learning by Doing'
On October 31, 2026, at the main forum of the 2026 AI+ Conference, Qingyan Technology (Beijing) Co., Ltd. and a consortium including The Chinese University of Hong Kong (Shenzhen), Beijing Institute of Technology, Shenzhen Hetao College, AIRS, Beijing Artificial Intelligence Association, and Jiuyuan Intelligent Computing System Ecosystem Alliance released the ReAct-Manip benchmark. This industry-first evaluation set shifts focus from single-task completion to continuous learning within tasks, assessing whether robots can improve performance through interaction feedback when facing unseen physical instances.
Simulation & World Models
Gazebo gains enhanced multi-robot simulation support through GSoC contributions
Open Robotics announced that Gazebo now features improved multi-robot simulation capabilities, including automatic discovery and creation of bridges for matching ROS topics and services, thanks to the Google Summer of Code (GSoC) work of Jiayi Cai. A discourse thread with demos is provided for further details.
Robots & Hardware
Atu Technology Launches 308,000 Yuan 'Robot Tailor' for Garment Manufacturing
Atu Technology has introduced the AiNi embodied intelligent sewing robot and the Production Line King, a mass-producible robotic production line. The AiNi robot, priced at 308,000 yuan in China, features automatic fabric recognition, dynamic wrinkle correction, and process self-learning via simple worker instruction. The Production Line King integrates an AI scheduling system with modules for cutting, distribution, sewing, and quality inspection. These products aim to address labor shortages, rising costs, and small-batch rapid response challenges in the garment industry.
Industry
White House Issues Proclamation On Robotics And Industrial Machinery: Top Points
On October 28, 2026, President Donald Trump issued a proclamation under Section 232 of the Trade Expansion Act of 1962 regarding robotics and industrial machinery imports.
Robotics & Automation NewsTutor Intelligence raises $75 million to scale robot workforce
Tutor Intelligence announced today that it has raised $75 million in Series B funding led by Andreessen Horowitz, bringing its total capital raised to $117 million.
Walmart Pumps $6B Into Robotics Partnership
Walmart has announced a $6 billion initial commitment to expand its partnership with Symbotic, bringing robotics automation to 25 U.S. regional distribution centers. The deal aims to improve e-commerce upstream inventory management, storage capacity, and product movement. Walmart is also retrofitting 42 of its regional distribution centers by 2030, with some sites like Opelousas, La., and Cullman, Ala., receiving over $330 million and $350 million respectively for modernization. Additionally, a new 744,000-square-foot facility near San Antonio, Texas, named "Project Theo," is set to begin construction in February 2027 and complete by August 2029, expected to create at least 425 jobs. Walmart's next-generation fulfillment centers use automated high-density storage and retrieval systems to reduce manual steps and double daily order processing capacity.
Pangqi Technology Lowers Embodied AI Robot Factory Entry Barriers via RaaS Model
Pangqi Technology, a company with over a decade of experience in intraplant logistics and more than 2,000 global projects, is deploying embodied AI robots through a Robot-as-a-Service (RaaS) model to reduce upfront hardware costs and operational risks. The approach addresses three engineering hurdles: environmental adaptation, multi-robot coordination via its proprietary ADS scheduling system, and cost/risk management. Pilot results from standardized warehousing scenarios show capacity increases up to 125% and order processing rates exceeding 15,000 per day, while production line pilots integrate humanoid robots into material handling workflows. The company plans to showcase its full-stack solution at CeMAT ASIA 2026 in Shanghai.
Briefs
- Microduck ROS 2 Integration: Running Policies in MuJoCo with ROS 2ROS Discourse — Community updates
- Topic statistics and libstatistics_collector removed from rclcpp in Rolling (Makoa Mata-mata)ROS Discourse — Community updates
- Integrating Low-Thrust Transfer Dynamics into POMDP Planning for On-Orbit ServicingarXiv Robotics — research abstracts
- Optimized fuzzy logic controller cuts average robotic arm tracking error by 89% in simulationsTech Xplore — Robotics
- ManuPrax Robotics Announces Nearly $10 Million in Funding for Dexterous Robot Hand SystemsChinese robotics — Research news discovery
- Sichuan Embodied Technology completes over 100 million yuan Pre-A round financing, accelerating humanoid robot mass productionChinese robotics — Company news discovery
- Xiaomi Robot Business Shows Dual Progress at 2026 World Internet ConferenceIT Home — Robotics
- Feijie Kesi Releases Fysiverse-UniWAM Unified World Action Model with Fudan University's CITLabChinese robotics — Research news discovery
- Shanghai Sazi Intelligent Technology Completes Pre-B Round FinancingChinese robotics — Company news discovery
- National AI Application Pilot Base (Embodied Intelligence) Plans Capital Increase and Share ReformRobotics — Chinese web discovery