Tracking these small, fast moving drones is something we’ve been working on recently: https://lnkd.in/g3WckMGm Our perception tech can work on… Tracking these small, fast moving drones is something we’ve been working on recently: https://lnkd.in/g3WckMGm Our perception tech can work on… In this paper, we present motion retargeting and control algorithms for teleoperated physical human-robot interaction (pHRI). We employ unilateral teleoperation in which a sensor-equipped operator interacts with a static object such as a mannequin to provide the motion and force references. The controller takes the references as well as current robot states and contact forces as input, and outputs the joint torques to track the operator's contact forces while preserving the expression and style… In this paper, we present motion retargeting and control algorithms for teleoperated physical human-robot interaction (pHRI). We employ unilateral teleoperation in which a sensor-equipped operator interacts with a static object such as a mannequin to provide the motion and force references. The controller takes the references as well as current robot states and contact forces as input, and outputs the joint torques to track the operator's contact forces while preserving the expression and style of the motion. We develop a hierarchical optimization scheme combined with a motion retargeting algorithm that resolves the discrepancy between the contact states of the operator and robot due to different kinematic parameters and body shapes. We demonstrate the controller performance on a dual-arm robot with soft skin and contact force sensors using pre-recorded human demonstrations of hugging. A broad perspective is presented for the utilization of modular robotic arms in various industrial tasks, particularly for cluttered environments. Modules are proposed to develop reconfigurable manipulators according to the robotic parameters, resulting out of the design procedure. A brief description of the modules divisions and the optimal assembly planning is presented. Focus of the paper is the multi-layered approach for modules inventory, which can be referred as base for the further… A broad perspective is presented for the utilization of modular robotic arms in various industrial tasks, particularly for cluttered environments. Modules are proposed to develop reconfigurable manipulators according to the robotic parameters, resulting out of the design procedure. A brief description of the modules divisions and the optimal assembly planning is presented. Focus of the paper is the multi-layered approach for modules inventory, which can be referred as base for the further additions in the types of modules required in the library. A case study on a realistic problem of challenging welding sites is presented as one example of the upper layer of multi-layered spectrum. The results present the designed configurations and corresponding modular assembly. Used NUMBA in Python to parallelize the operations of Gaussian Filtering, Fast Fourier Transform and PSD on an Audio Wave. The temporal aspects for a spectrogram so generated were conserved up to a much higher resolution. Useful results and inferences on effect of data transfer between GPU and a CPU onto computation time were drawn Batch Normalization and dropout layers are used in the 10 layered Convolutional Neural Network architecture. Keras is used for all experiments with a Tensor Flow Backend. Successfully implemented Adam Optimizer with learning rate of 1e-3 and no decay rate. Achieved 7.5 degree of mean error in angle approximation. A generalised MPC based iLQG Controller was implemented using Python and MuJoCo which successfully drove an under actuated cart pendulum from any initial position (including swing up) to the final upright position. Algorithm successfully tested on Half - Cheetah and Point Mass Environments as well. Regularization Scheduling was added to make sure that the quadratic cost approximation has a convex minimum. Undergraduate Thesis : A cost-effective Haptic based master-slave robotic system that can extend an expert doctor's diagnostic treatment to poor patients in rural areas.

A 6 degree of freedom serial open chain manipulator was modeled in Solid Works. The kinematic constraints were inferred from robot's required dexterous work space generated in MATLAB. All the robotic links were designed sequentially from last link to the first such that entire assembly's weight is distributed optimally.… Undergraduate Thesis : A cost-effective Haptic based master-slave robotic system that can extend an expert doctor's diagnostic treatment to poor patients in rural areas.

A 6 degree of freedom serial open chain manipulator was modeled in Solid Works. The kinematic constraints were inferred from robot's required dexterous work space generated in MATLAB. All the robotic links were designed sequentially from last link to the first such that entire assembly's weight is distributed optimally. Finite Element Analysis was carried out on complex sections using ANSYS.

A model based controller for single link master-slave system was developed on MATLAB/SimuLink. Haptic force signals were successfully recorded while simulating above system's model using ODE45-solver. A MATLAB algorithm is proposed which generates dynamic models for model-based control of reconfigurable modular manipulators. The system model evaluated for original prototype was used with an Automated Computed Torque Controller designed in SimuLink.

The scheme was then validated by simulating trajectories on a SolidWorks Assembly with negligible error, using ODE 45-Solver and Sim-Mechanics. The idea of 'Multi-Layer Approach in Robot Modularity' was proposed, which divides the large spectrum applications of a single modular robot library into a multi-level hierarchy for efficient adaptability towards different tasks.

A case study on application of modular manipulators in repair & maintenance welding has been showcased. 🚀 2023: A Landmark year for Apptronik as we unveiled Apollo, a robot that embodies our enhanced focus and mission. 👨‍🚀 2024: Big things are in… 🚀 2023: A Landmark year for Apptronik as we unveiled Apollo, a robot that embodies our enhanced focus and mission. 👨‍🚀 2024: Big things are in… A recent post talked about mesh distortion affecting displacement results. However it looked very much like direct nodal forces had been applied to the edge of the rectangular plate used. These dominated the results. If you define a constant distributed edge load in a preprocessor it calculates a set of equivalent nodal forces. These nodal forces effectively take into account the stiffness of the Degrees of Freedom at each node and make sure the actual force applied creates a kinematic balance. The right hand model shows the classic 'corner curl' ; there is a full nodal force (2 units) applied at the edge nodes which only have a stiffness contribution of one element. This is in contrast to the interior nodes which have contribution from two elements. Hence double the deflection at the corners! The left hand model shows the correct nodal force distribution created by defining an edge loading. I have also introduced varying element lengths as in the original post, just to make it more interesting. You can find out the actual nodal forces by using a freebody tool as I did. FEMAP also has an interesting option to convert a distributed load, or pressure, to equivalent nodal forces - but it's one way only! The bottom line is - don't try to second guess the nodal forces, particularly with higher order elements and non-uniform load distributions. AND DON'T TRUST SUSPICIOUS RESULTS! The autonomous vehicle industry is at a crossroads. Waymo's co-CEO Tekedra Mawakana recently warned that robotaxi companies must go beyond marketing and deliver concrete safety proofs to regulators and the public. With recent incidents involving competitors like Cruise, the pressure is on to establish standardized safety metrics and transparent validation processes. Key challenges include: 1. Data transparency: Companies must share safety test results and incident reports openly. 2. Human-in-the-loop verification: Autonomous systems need rigorous human oversight during edge-case scenarios. 3. Regulatory alignment: Collaborating with agencies to create measurable safety benchmarks. For tech founders building mobility solutions, here's what to prioritize: - Invest in real-world safety testing with third-party validation. - Develop public dashboards to track safety KPIs in real time. - Advocate for industry-wide safety protocols that regulators can audit. The path to mass adoption isn't just about innovation—it's about trust. How will your team balance speed with safety in autonomous tech development? Let's discuss. 🚍 What if shifting a bus departure by just a few minutes could save an entire vehicle? That’s the idea behind our new paper, recently published in the INFORMS Journal on Computing. Together with Thomas van der Schaft, we developed a Dynamic Discretization Discovery (DDD) algorithm that tackles the Multi-Depot Vehicle Scheduling Problem with Trip Shifting (MDVSP-TS). Why this matters: - Even tiny timetable adjustments can unlock huge efficiency gains in public transport. - Our algorithm solves real-world bus networks with nearly 4,000 trips to optimality. - It outperforms existing exact methods by a wide margin. This work shows how smart optimization can make public transport more efficient, sustainable, and practical at scale. 👉 Read more here: https://lnkd.in/eS_dhZZ3 A new depth estimation model for Robotics Depth Anything 3 (DA3) just dropped, and it’s a great step forward for perception in robotics. Congrats to the authors Haotong Lin, Sili Chen, Jun Hao Liew, Donny Y. Chen, Zhenyu Li, Guang Shi, Jiashi Feng, and project lead Bingyi Kang for this impressive open-source release. Why it is interesting for robotics: ✅ Much better monocular depth ✅ Consistent multi-view geometry → great for mapping & SLAM ✅ Pose estimation from images ✅ Metric depth when scale matters ✅ Foundation models that work across many vision tasks 🔗 If you’re building perception or navigation pipelines, DA3 is definitely worth a look. Which perception stack are you using in your Robotics projects? Let's connect and share Robotics resources 🔽 #Robotics #AI Touring RLSO 4: Stochastic gradient methods as a sequential decision problem Continuing my tour of “Reinforcement Learning and Stochastic Optimization” … Chapter 5 of my RLSO book shows that any stochastic gradient algorithm can be written as a sequential decision problem. The graphic below shows the standard stochastic gradient (for a maximization problem) using a simple Kesten stepsize rule (but this works for any stepsize rule). The rest of the graphic shows the five elements of the universal modeling framework for an SDA using this update. Note that a “stepsize rule” is a form of policy function approximation (PFA), which is just one of the four classes of policies. Note that all PFAs are parameterized functions that need to be tuned, and the tuning depends on the initial state S_0 (which includes, among other things, the choice of starting point).  If this were a deterministic nonlinear programming problem, we would use a one-dimensional search, which his a form of direct lookahead approximation (DLA). A potential path for overcoming the need for tuning (which is serious, and often overlooked) would be to devise a stochastic lookahead. Most important is the realization that we should at least think about the other three classes of policies. Or perhaps we should look into CFAs or VFAs. Fresh thinking about a very old problem. AI for Science is shifting the paradigm in Computational Fluid Dynamics (CFD)—moving us from hours of simulation to seconds of inference. The Solution Architect and PhysicsNeMo team (Abouzar Ghasemi Ira Shokar Farah Hariri Mohammad Amin Nabian Ram Cherukuri) at NVIDIA have done an awesome in developing a learning kit to bridge the gap between AI and CFD. If you are looking to get hands-on with AI Physics, here is a complete walkthrough using the canonical Ahmed Body benchmark. These two Jupyter notebooks guide you through the lifecycle of a deep learning surrogate model using the PhysicsNeMo DoMINO (Decomposable Multi-scale Iterative Neural Operator) architecture. Here is the breakdown: Notebook 1: Data Preprocessing Raw simulation data needs to be ML-ready. We cover the extraction and transformation of the Ahmed Body Surface dataset, tackling geometry handling, point cloud generation, and preparing inputs for the model.  Explore it here: https://lnkd.in/epM_TZGJ Notebook 2: Training the Model We dive straight into configuring and training the DoMINO architecture—specifically designed to handle the multi-scale challenges of 3D aerodynamic flows. You will learn how to set up the training loop, handle loss functions, and validate performance on unseen geometries.  Explore it here: https://lnkd.in/etMTjpWq This workflow is a prime example of End-to-End AI for Science, bridging the gap between traditional CAE data and modern neural operators. Explore the full repository:  https://lnkd.in/eG_div-D 🚗 Are AVs Finally Just Around the Corner? When I started my PhD in 2016, AVs were “just around the corner”. Nearly a decade later, they still are—but the landscape is shifting… 🔥 Xpeng’s CEO says AVs are on the verge of a “ChatGPT moment”, meanwhile test drives still require driver intervention. 🚕 While China’s robotaxi industry has over 2,300 driverless cabs operating in limited areas—far ahead of the U.S. with ~700 — they remain unprofitable. ⚡ ADAS competition: BYD is adding self-driving features to $10,000 cars, while Tesla charges $8,800 for similar tech. 🏛️ Regulatory roadblocks: China is pushing for level 3 autonomy, but safety concerns and liability questions remain. 💸 GM pulls back: After sinking $10bn into robotaxis, GM axed its program, shifting focus to assisted driving instead. 📉 Lidar costs plummet: Once thousands of dollars per unit, lidar sensors now cost around $200, making AV tech far more affordable. 🔥 The unexpected ChatGPT revolution has certainly revived hopes that AI could finally overcome AVs biggest hurdles. 💬 What do you think, are AVs really just around the corner and if so is this going to cause a major shake up the car industry between leaders and laggers? Read further in this interesting Economist article (link in comments). The Sapper SMET / RCV size category robotic platform, built by Applied Research Associates, at AUSA 2025. One of several robotic platforms presented at the conference that makes me genuinely wonder why so many robotic vehicle designers are using lots of tiny rollers in a largely rigid suspension mount, for what should be relatively fast, relatively large, cross-country tracked platforms - especially since for manned tracked combat vehicles, this track configuration hasn't been used since the Churchill Infantry Tank of World War 2 (and even then, it was an outdated suspension design) Last month, I had the opportunity to present at the MathWorks Automotive Conference alongside the RoadRunner team, Kunal Patil, Ph.D., showcasing how GeoMate’s simulation-ready maps integrate seamlessly with RoadRunner to accelerate AD/ADAS development. In this short presentation, I walk you through our approach to scalable and cost-effective mapping for virtual testing environments. 🎥 Watch the video below 📄 Full slide deck available in the comments #ADAS #AutonomousDriving #Simulation #HDMaps #MathWorks #RoadRunner #GeoMate Diffusion policies forget the past. This method teaches robots to remember… with 3x better performance ⬇️ Robots need memory to act reliably over time, but adding history to diffusion-based policies usually makes performance worse and training more expensive. This new work introduces PTP, an auxiliary loss that brings back context understanding and makes long-history learning fast and effective. Why it matters ✅ PTP helps diffusion policies learn meaningful past-future connections ✅ Enables efficient training with cached short-context embeddings ✅ Boosts performance by 3x while cutting compute costs by over 10x ✅ Adds a test-time scaling trick that checks if the robot is paying attention to its own history Learning long-context robot policies just got way more practical. If we want smarter, more capable robots, this is a big step in the right direction. Thank you so much for sharing this, Marcel Torne Villasevil 🙏 Check out the paper and real-world tasks solved with PTP: Paper: arxiv.org/abs/2505.09561 Website: long-context-dp.github.io Code: https://lnkd.in/dPK2uG9x