Applied EV - Software Defined Machines™’s Post

We are excited to announce that our paper, "Collaborative Multi-Object Tracking with Conformal Uncertainty Propagation," has been accepted by the IEEE Robotics and Automation Letters! This is a huge achievement for our team, and we are grateful to the review committee for recognizing the importance of our work. Object detection and multiple object tracking (MOT) are crucial for autonomous vehicles, ensuring safety and robustness. Our uncertainty propagation framework, MOT-CUP, is the first attempt to address the challenges of leveraging the uncertainty quantification from collaborative object detection (COD) to enhance MOT performance. Our framework quantifies the uncertainty of COD through direct modeling and conformal prediction, and propagates this information into the motion prediction and association steps of MOT, resulting in a remarkable 2% improvement in accuracy and a 2.67x reduction in uncertainty compared to baseline algorithms like SORT and ByteTrack. Moreover, in high occlusion scenarios, MOT-CUP shines with a 4.01% accuracy enhancement, making it a champion in occlusion. MOT-CUP demonstrates the importance of uncertainty quantification in both COD and MOT, and provides a significant improvement in accuracy and uncertainty reduction in MOT through uncertainty propagation. Check out our website and paper to learn more about MOT-CUP and our efforts to improve the safety and performance of autonomous driving. #SelfDriving #ObjectDetection #MOT #UncertaintyQuantification

One of the modules we dropped during the break is a really big deal - Follow-the-Leader (or, Follow with an Offset). You can use it for a variety of functions, like having an autonomous vehicle follow a manned vehicle along a pre-set path, or having a tractor pulling trailers follow a harvester at an offset. Commands that we make possible through a simple API command, so you can spend your engineering effort not on being the 50th team to build Follow-the-Leader, but on the integration with your harvest ERP (or your defense system for platooning!).

Discover how Gaussian Splatting revolutionizes AV simulation! This cutting-edge technique enhances the realism of autonomous vehicle simulations by efficiently rendering complex environments with high fidelity. Learn how Gaussian Splatting can improve safety and performance in autonomous driving technology. #gaussiansplatting #avsimulation #autonomousvehicles #selfdrivingcars #simulation #automotive #machinelearning #deeplearning #computervision #technology #rendering #highfidelity #complexenvironments #cuttingedge #futureoftransportation #autonomousdriving #ai #artificialintelligence #vehicletechnology #simulationtechnology

Excited to announce that our research paper, "Navigating Unsignalized Intersections: A #Predictive Approach for #Safe and #Cautious #Autonomous_Driving," has been published in IEEE Transactions on Intelligent Vehicles. This work addresses the challenging problem of online #speed_planning along a predefined traverse path based on an interaction-aware #prediction_algorithm in an unsignalized intersection. Our goal is to mimic the behavior of a real driver and calculate the appropriate acceleration value at every time step based on that behavior. The algorithm aims to promote a more comprehensive and #cautious_driving_behavior in autonomous vehicles. I invite you all to read our peer-reviewed version which can be accessed through the following link: https://lnkd.in/dnQ6Z8S5 Special thanks to my co-authors, Amir Ghafari and Prof. Ali Ghaffari (K. N. Toosi University of Technology). #Autonomous_vehicles #Automobiles #Behavior_prediction #Interactive_behavior #Unsignalized_intersection #Graph_neural_network

Real-time trajectory planning for autonomous driving harnesses the strengths of Gaussian Processes (GP) and Incremental Refinement to navigate dynamic environments safely and efficiently. Gaussian Processes offer a probabilistic framework for modeling uncertainties in vehicle dynamics and environmental conditions, ensuring that the planned trajectories are smooth and have a high confidence level. This is crucial for maintaining safety and comfort during autonomous driving. Incremental Refinement complements this by providing a mechanism to iteratively improve the trajectory in real-time, adapting to new data and environmental changes as they occur. This iterative process allows for continuous adjustments, ensuring that the vehicle can respond promptly to unexpected obstacles or alterations in the driving scenario. By integrating the probabilistic modeling capabilities of GPs with the adaptive nature of Incremental Refinement, autonomous driving systems can achieve a high degree of reliability and responsiveness, essential for real-world operation. The video is for the paper “Real-Time Trajectory Planning for Autonomous Driving with Gaussian Process and Incremental Refinement”, accepted by ICRA2022. Authors: Jie Cheng, Yingbing Chen, Qingwen Zhang u Gan and Ming Liu Paper: https://lnkd.in/gGD3ktmy Github: https://lnkd.in/gT2DSZ25 #autonomusdriving #algorithm #ros #carla #robotics

💥 Holiday module drop 👇 This week we're dropping 2 more modules designed to radically simplify off-highway autonomy. FOLLOW-THE-LEADER MODULE: Some vehicles work better in pairs (think harvesting, for example). But orchestrating a vehicle to follow a moving target is a complex problem. That's where our Follow-the-Leader module comes in. You can easily direct one vehicle to follow another at a specified offset. SIM-TO-REALITY PIPELINE MODULE: Most robotics simulations don't accurately simulate vehicle kinematics and dynamics, meaning they don't actually help the development process that much. Our Sim-to-Reality Pipeline module bridges the gap by focusing on the vehicle's motion profile (rather than on pretty visuals). The result? Behaviors and tuning developed in sim can finally be used on your physical vehicle. Want to learn more about our autonomous navigation modules? Read up on the Polymath blog: polymathrobotics.com/blog

🤔 What are the significant challenges in detecting and reacting to unexpected objects on the road? We explore how processing information accurately and determining the safest course of action can mitigate these challenges! 🚙 Join Intellias, Câmara Municipal de Lisboa, Bosch Portugal and Critical TechWorks as they discuss the advances of autonomous driving. Catch the full conversation here 👉 https://bit.ly/3xnOYhI #JoyInMotion #CriticalTechWorks #BMWGroup #Agile #Software #Engineering #CTWTalks

In this paper, delve into the challenges of ensuring stability (in some sense) and robustness in large-scale second-order consensus networks (SOCNs) and autonomous vehicle platoons in the discrete-time domain. It proposes a graph-theoretic methodology for designing a state feedback law for these systems in a discrete-time framework and concludes with simulation and experimental results (using Quanser's QLabs and QCars) that show that the proposed framework is very effective in predicting the behaviour of vehicle platoons, even for more complex agents with nonlinear dynamics. Read now! https://lnkd.in/dGSzD_xC #robustness #stabilityanalysis #graphtheory #trajectory #gaussiannoise #vehicledynamics #northeasternuniversity

🔥 Read our Highly Cited Paper 1. Mechanism and Application of Layered Grouting Reinforcement for Fractured Coal and Rock Roadway 🔗 https://lnkd.in/gqZRm3AF 2. Regenerative Braking Control Strategy Based on AI Algorithm to Improve Driving Comfort of Autonomous Vehicles 🔗 https://lnkd.in/gJ2qrf8K 3. A Modified Folin-Ciocalteu Assay for the Determination of Total Phenolics Content in Honey 🔗 https://lnkd.in/gxamAX6S