I will be attending ICRA this year. Shoot me an e-mail if you are too and would like to join a small WR get-together or catch-up during the event. We are kicking off today’s issue with a humongous amount of news related to simulators, as my backlog started going out of proportion. As usual, the publication of the week section is manned by Rodrigo.
Weekly Robotics is being developed thanks to the Patreon supporters and the following business sponsors:
Simulators, simulators everywhere
MultiVehicle simulator is a “lightweight, realistic dynamical simulator for 2D (“2.5D”) vehicles and robots. It is tailored to the analysis of vehicle dynamics, wheel-ground contact forces and accurate simulation of typical robot sensors (e.g. 2D and 3D lidars)”.
CoppeliaSim V4.5 was released last month. For more information about the added functionalities, see this ROS Discourse post or this video showing kinematics debugging.
In the autonomous car world (and tangents), AVL engineers added a capability of off-road driving to CARLA simulator. Adding a soft-soil tire model allows you to drive tractors in CARLA as soon as the changes are merged upstream.
VRX Simulation Environment v.2.1.0. This simulator is closely tied with the robotX challenge and specifically the VRX Competition. The simulator is based on Gazebo and supports all the tasks the competition requires.
Lastly, Aditya Kamath recently wrote an article on “Visualizing Robots in Unity”. It’s not strictly a simulator, as Aditya only visualizes the robot’s state, but it’s another tool for your tool belt.
Little Robots Learn to Drive Fast in the Real World
This article introduces FastRLAP, a system for high-speed driving via Deep Reinforcement Learning and Autonomous Practicing. Using a pre-trained “foundation model”, the robot can teach itself to race around tracks in 20 minutes. The exciting feature of the learned policy for outdoor driving is that the model can distinguish ground characteristics and opt for high-traction areas. For more information about this project, check out the project’s website
Auto Tape Wrapping Machine Is Amazing For Cable Management
I know, I know, no sensor - not a robot, but I enjoyed this video by The Q, where he builds a simple wire-looming machine. I appreciate the video montage and the ASMR vibes.
Learning Agile Soccer Skills for a Bipedal Robot with Deep Reinforcement Learning
This video popped into my feed the other day. In this research from DeepMind, the team used Deep Reinforced Learning to teach Robotis OP3 small humanoid robots with 20 DOF and motion capture vests to play soccer. I have to admit, and I’m not exaggerating, that I’m mind blown by how these small robots perform.
World’s First Fully Remote Offshore Wind Farm Inspection Completed in the UK
Fugro’s Blue Essence USV run a fully-remote inspection of Wind Turbines in Aberdeen. As a part of the mission, the system delivered information about the structure of wind turbines and a detailed map of the seabed obtained using a multibeam echosounder.
Publication of the Week - Multi-Camera Visual-Inertial Simultaneous Localization and Mapping for Autonomous Valet Parking (2023)
I can’t wait until I don’t have to find a spot to park my car and let the car choose for me. This paper presents a visual-inertial SLAM (Simultaneous Localization and Mapping) for autonomous valet parking using four monocular fisheye cameras. The method uses a fast segmentation network, loop closure detection, and correction to generate a 3D reconstruction of the space. The authors tested the system on simulation and real-world datasets from Ford Motor Company, and results show an improvement over state-of-the-art methods, such as ORBSLAM3. You can see the system in action this video.
Robotics safety firm Veo raises $29 million, with help from Amazon
Veo Robotics, a company working on Speed & Separation Monitoring (SSM) for work cells, raised a $29M series B. Congrats!