AI & Learning
Reinforcement Learning · Learning-Based Control · Imitation Learning · Deep Learning
Autonomous Systems · Robotics · AI
Engineeringintelligencefor systems thatmove.
I'm Wenliang Zhang, a researcher and engineer working on motion planning, control, simulation, and machine learning for autonomous systems.
10years in motion planning, control, and simulation for autonomous systems
13peer-reviewed publications with 300+ citations
20+M.Sc. students supervised in industry-linked projects
116k+GitHub downloads across my Obsidian plugins
Skills
Methods across the engineering loop.
Optimisation-based planning and control, state estimation, simulation, and learning-based methods for safety-critical autonomous systems.
Core DomainAutonomous
Systems
Systems
Planning & Control
Motion Planning · Model Predictive Control · Trajectory Tracking · State Estimation
Robotics & Autonomy
Autonomous Driving · Robotics · Decision-Making · Safety-Critical Systems
Simulation & Engineering
Physical Modelling · Digital Validation · Co-Simulation · Real-Time Optimisation
Model → Plan → Control → Learn → Validate → Model
A closed-loop workflow: validation feeds back into models, objectives, and assumptions.
Model
Physical and learned representations at the fidelity required for prediction, design, and real-time decisions.
Modelica · Dymola · CarSim / TruckSim · CarMakerPlan
Motion, trajectory, and decision problems formulated with safety, progress, energy, and comfort.
CasADi · acados · rockit · Gurobi · IpoptControl
Constraint-aware control and state estimation that turn planned motion into closed-loop behaviour.
MPC · OCP · MHE · EKF / UKFLearn
Reinforcement learning that complements physical structure and improves adaptation.
RL · DL · Learning Control · leap-c · PyTorchValidate
Repeatable scenarios and evidence that expose closed-loop strengths, limits, and trade-offs.
Scenario Design · Co-Simulation · FMI / FMU · PythonNonlinear MPCTrajectory PlanningCasADiWheel-Slip ControlCombined-Slip ModellingModel-Based RLDynamic ModelNeural Tyre ResidualsTrajectory OptimisationTorque VectoringMPCCOCPNeural NetworksTrajectory GenerationTransformer-Based ModelsAutonomous RacingModelicaHPIPMModel ValidationYaw Stability ControlDriving Simulator StudiesLinear Tyre ModelsSteer-by-WirePyTorchState EstimationClosed-Loop EvaluationAgent Behaviour ModellingSingle-Track ModelAgentic CodingFMI / FMUState EstimationReinforcement LearningRobust / Adaptive ControlImitation LearningPath FollowingIpoptEKF / UKFSafety GuaranteesInterior-Point MethodsSafety-Constrained ControlFour-Wheel SteeringRide & HandlingPythonMCP ServersCarSim / TruckSimLoad Transfer AnalysisTyre Parameter IdentificationBehaviour PlanningKinematic ModelHPMPCMPCToolsSignal ProcessingEmbedded OptimisationPhysics-Informed Data GenerationAutonomous SystemsScenario Designleap-cscikit-learnSIL / HIL TestingNumerical OptimisationAutomated Evaluation WorkflowsAutonomous DrivingFault ToleranceNumPyODE / DAEMATLAB / SimulinkOver-Actuated VehiclesMulti-Objective OptimisationCollision AvoidanceABS / ESC / TCSrockitActive CamberElectric VehiclesReal-Time OptimisationReal-Time Decision MakingClaude Code / CodexMotion PlanningMHERoboticsModel EvaluationLearning-Based ControlMulti-Agent OrchestrationPrompt EngineeringMotion ComfortYALMIPqpOASESBrush Tyre ModelSim-to-Real ValidationLimit-Handling ManoeuvresDouble-Track ModelTyre ModellingGurobiSideslip ControlQuadratic ProgrammingEnergy-Efficient ControlMultibody DynamicsSensor FusionMPCOptimal ControlNonlinear ProgrammingMagic Formula Tyre ModelBehaviour CloningCarMakerActive SafetyEnd-to-End LearningSafety-Critical SystemsVehicle DynamicsDymolaacadosDigital ValidationFMU-Based Co-SimulationReward ShapingDeep LearningTrajectory Tracking
Projects
Selected projects.
Problems, methods, and results from current and completed projects.
A
Current Research Direction
Learning-based autonomy
Combining model predictive control with reinforcement learning for motion planning and control—learned policies under safety constraints, with real-time feasibility and interpretability as design requirements.
Reinforcement LearningDeep LearningModel Predictive ControlAutonomous Systems
B
Research Project · 2026 — 2027
Robust control and safety guarantees
Contributing to a Chalmers research project on robust control and safety guarantees for over-actuated autonomous electric vehicles—failsafe behaviour and control allocation under degraded actuators.
RobustnessSafety GuaranteesFault ToleranceOver Actuation
C
Research Project · 2023 — 2025
Multi-objective motion planning and control
Principal investigator at KTH for an industry-linked project on multi-criteria motion control of automated electric vehicles—treating safety, energy, comfort, and travel time as coupled design objectives, validated on a test vehicle.
Multi-ObjectiveSafetyEnergy EfficiencyMotion Comfort
D
Engineering Infrastructure
Simulation and validation
Built dynamic models and FMU-based co-simulation workflows—closed-loop evaluation of models, algorithms and scenarios before physical testing.
Dynamic ModellingFMUScenario DesignClosed Loop
Demos
Control in motion.
Videos from an over-actuated autonomous electric vehicle platform, showing how vehicle dynamics, optimisation, and learning-based control behave in daily driving and handling limits.
Generative roads for closed-loop testing
Open-ended and closed-circuit roads are generated in the same representation the MPC reads: curvature, width, slope, banking, and friction. Rule-based layouts can be refined by a VAE or diffusion model. The road extends only slightly beyond the prediction horizon for long-running closed-loop and preview testing.
View Post and Discussion on LinkedInMPC in 5 manoeuvres with 3D body motion
Double lane change, slalom, steady-state circular, step steer, and accelerate-then-brake, each driven closed-loop by an MPC controller. Per-wheel suspension makes the body roll, pitch, and heave visible, and the physics, control, 3D rendering, and tyre sound all run on open-source Python packages.
View Post and Discussion on LinkedInMPC-controlled drift on a 50-metre circle
The MPC holds a steady-state drift for a 50-metre circle at roughly 19° sideslip, with visible counter-steering in a 3D model with per-wheel suspension.
View Post and Discussion on LinkedInReinforcement learning tunes the MPC driver
Fields of 50 and 100 learned drivers—each car a checkpoint from one reinforcement learning run—show state-dependent MPC weights, ranked by lap time from still-learning to polished performance.
View Post and Discussion on LinkedIn18 MPC drivers from Optuna auto tuning
Optuna uses Bayesian optimisation to tune 18 fixed MPC configurations, creating distinct controller personalities that trade off grip, rotation, and lap time through torque vectoring.
View Post and Discussion on LinkedInExperience
From a PhD in vehicle engineering to learning-based autonomy.
Research, teaching, and supervision roles, with multi-partner projects throughout.
Postdoc Researcher · Systems and Control
Chalmers University of Technology, Gothenburg, Sweden
Motion planning and control combining MPC with RL under safety constraints. Co-supervising a PhD student and a M.Sc. thesis on learning-based control; supervised a nine-student bachelor's project that built a small-scale over-actuated car, from hardware to algorithm.
Principal Investigator & Research Lead
KTH Royal Institute of Technology, Stockholm, Sweden
Led multi-partner research on multi-criteria motion control, energy-efficient trajectory planning, and motion-comfort evaluation.
Postdoctoral Researcher
KTH Royal Institute of Technology, Stockholm, Sweden
Developed planning, control, and simulation methods for over-actuated autonomous electric vehicles; taught the M.Sc. course Applied Vehicle Dynamics Control and supervised industry-proposed thesis and project teams.
Technical Reviewer
IEEE Transactions and international journals
Reviewed 30+ papers on autonomous driving, vehicle control, and learning-based methods for venues including IEEE Transactions on Vehicular Technology and Vehicle System Dynamics.
PhD · Vehicle Engineering (Autonomous Driving)
KTH Royal Institute of Technology, Stockholm, Sweden
Research on vehicle dynamics, active safety, optimal control, state estimation, and FMU-based co-simulation for over-actuated autonomous electric vehicles.
Contributions
Research, talks, code, and writing.
Peer-reviewed papers and invited talks from the research; open-source tools and articles from the engineering and writing practice alongside it.
Selected ResearchFull Record
2026Conference PaperExploring a learning-based approach to motion planning and control for over-actuated autonomous vehiclesAVEC 2026
2025Conference PaperExploring motion comfort and time optimal trajectory planning in autonomous vehicles through over-actuationIAVSD 2025
2024Conference PaperEnergy and time optimal control of autonomous vehicles by using Frenet frame modelling and over-actuationAVEC 20242022Doctoral DissertationExploiting over-actuation for improved active safety of autonomous electric vehiclesKTH Royal Institute of Technology2021Conference PaperExploring model complexity for trajectory planning of autonomous vehicles in critical driving scenariosIAVSD 20212020Journal Article · 100+ CitationsEvaluating model predictive path following and yaw stability controllers for over-actuated autonomous electric vehiclesIEEE Transactions on Vehicular Technology2019Journal ArticleAdvanced vehicle state monitoring: Evaluating moving horizon estimators and unscented Kalman filterIEEE Transactions on Vehicular TechnologySelected Talks
- 2025-05FMU-based co-simulation for autonomous vehicle dynamics controlSVEA Vehicle Dynamics Seminar
- 2024-09Multi-criteria optimal motion control of automated EVsSwedish Electromobility Centre Thematic Group Meeting
- 2023-05Exploiting over-actuation for improved active safety and energy efficiency of autonomous electric vehiclesSyscop Seminar · University of Freiburg
- 2022Exploiting over-actuation for improved active safety of autonomous electric vehiclesInvited Research Talks in Academia and Industry
- 2022-05Active safety of autonomous electric vehiclesSVEA Vehicle Dynamics Seminar
Tools for ThinkingGitHub
Task Marker
Change task status and append timestamped text for context.
Workflow State · Interaction DesignTodoist Context Bridge
Bridge notes with Todoist tasks while preserving the rich context tasks came from.
Context Preservation · IntegrationTag Index
Curate a list of important tags as meaningful entry points to a knowledge base.
Information Retrieval · CurationSend to Canvas
Send tasks, blocks, and notes to canvas files as plain text, links, and embeds.
Visual Thinking · Information ArchitectureLink Maintainer
Maintain note links automatically when splitting or reorganising notes.
Change Safety · RefactoringSelected from 12 open-source Obsidian plugins I developed for PTKM.
Selected ArticlesAll Articles
- 2025-07How to manage 240+ projects without burnout: Long-term strategies from my 8-year journeyProject Portfolio Strategy
- 2025-07Two thinking systems — how to use them for complex challengesThinking Methodology
- 2025-06Task switching, attention residue, and state resetAttention and Focus
- 2025-05The strategy of note taking: Redundancy, reliability, efficient discovery, and beyondKnowledge System Design
Ongoing
Ongoing work.
Safety-assured control, learning-based method, scenario generation, and reproducible simulation.
- Researching motion planning and control that combines MPC with reinforcement learning
- Advancing robust control and safety guarantees for over-actuated autonomous electric vehicles
- Exploring virtual road generation for scenario-rich generalisation
- Improving reproducible simulation and real-time optimisation workflows
- Building systems and tools that connect ideas, execution, and durable knowledge