Autonomous Driving and Planning
End-to-end planning, latent planning states, perception-to-control pipelines, and controller design for self-driving systems.
Autonomous Driving · Robotics · KAIST AXE Lab
Satyam Mishra
PhD Researcher at KAIST · AI for machines that perceive, plan and act
I build AI for autonomous vehicles and robots. My PhD research focuses on end-to-end planning for self-driving cars, and beyond the car I work on physical AI: robot perception, simulation, and reinforcement learning that turns plans into safe, reliable control.
143
Citations
7
h-index
6
i10-index
38+
Research outputs
6+
Years in AI
About
Research identity
I am an AI researcher and engineer at the KAIST Kim Jaechul Graduate School of AI, conducting PhD research in the Automotive Intelligence X Electrification (AXE) Lab. My research focuses on end-to-end planning for autonomous driving: what planning modules represent in their latent states, and the control architectures that turn those plans into motion.
The thread through my work is intelligence that holds up outside the lab: systems that are interpretable, stable, and safe when they meet the real world. That runs from self-driving cars to robots and physical AI, through reinforcement learning and safe control, to tools for understanding what large models have learned.
Alongside my core research, I build complete robotics systems that span perception, planning, control, and simulation. I also serve the research community as a NeurIPS 2026 reviewer and an AAAI 2026 Program Committee member.
Autonomous Driving
Robotics
Physical AI
Planning and Perception
Reinforcement Learning
Safe Control
Robot Simulation
Interpretable AI
NLP and LLMs
Research
Research directions
Four connected directions, all aimed at AI that perceives, decides, and acts reliably in the physical world.
Reinforcement Learning and Safe Control
Safe and explainable RL, constrained decision-making, and symbolic oversight for policies that stay predictable on real systems.
Interpretable AI and Model Dynamics
Koopman operators, spectral methods, and entropy signals for understanding learned models, from planning latents to semantic drift and hallucination in LLMs.
Robotics and Physical AI
Robot perception, mobile robots, motion planning, and simulation for embodied systems that close the loop from sensing to control.
- L1Perceptionsensors · detection · BEV
- L2Predictionagents · motion · intent
- L3Planningend-to-end · latent states
- L4Controltrajectory · actuation
Autonomy stack
From sensors to steering
My PhD research at KAIST AXE Lab focuses on end-to-end planning for autonomous driving: how planners represent the world in their latent states, and how those representations connect to control architectures for self-driving systems.
- L1Perceiving the scene from cameras and LiDAR
- L2Anticipating how surrounding agents will move
- L3Planning safe, interpretable trajectories
- L4Turning plans into stable vehicle control
Robotics · Physical AI
Intelligence that moves
A self-driving car is one kind of robot. Beyond the car, I work on physical AI more broadly: robots that perceive their surroundings, are developed and tested in simulation, and act safely in the real world. It is the same perception, planning, and control loop as my driving research, applied to a wider family of machines.
Mobile robots
From a Jetson Nano self-driving robot car with deep-learning collision avoidance and path following to complete perception, planning, and control stacks.
Robot perception
Detection, measurement, pose estimation, and 3D point clouds that turn sensor data into geometry a robot can act on.
Motion, simulation and learning
Koopman-based one-shot motion planning, simulation, and safe reinforcement learning to shape robot behavior before it reaches hardware.
Closing the loop
Connecting perception, planning, and control into complete robotic systems that behave reliably outside the lab.
Publications
Full publications and manuscripts
Citation counts and metrics from Google Scholar, updated October 2026.
Projects
Selected applied AI systems
Autonomous Robot with Jetson Nano
Developed a self-driving robot car with deep learning for collision avoidance, path following, and image recognition.
Research-Driven AI Agent
Built a research-grade AI agent using Gemini, LangChain, tool calling, DuckDuckGo search, and Pydantic-based structured outputs.
View on GitHub →SATMeas Object Detection and Measurement
Built a real-time object detection and dimension measurement system using computer vision, reaching over 98% reported accuracy.
Conversational AI with MistralAI
Created a Streamlit chatbot using Mistral-7B-v0.1, with practical deployment considerations for generative AI applications.
Live demo →Email Sentiment Analysis
Built a platform for analyzing email sentiment with Hugging Face Transformers, NLTK, and language-model-based response generation.
Image Captioning Web App
Deployed an image captioning app with Transformers, PyTorch, and Pillow behind a Streamlit interface.
Live demo →Patents
Inventive work
Books
Books and learning resources
Recognition
Awards and academic service
Reviewer, NeurIPS 2026
Serving as a reviewer for the Conference on Neural Information Processing Systems (NeurIPS 2026).
Program Committee, AAAI 2026
Serving as a Program Committee member for the AAAI Conference on Artificial Intelligence (AAAI 2026).
Paper presented at IEEE ICDM 2025 (A*)
Koopman Spectral Dynamics for Interpretable and Predictive Semantic Drift in Language Models.
Valedictorian Award
Vietnam National University, Hanoi.
Second Prize and Best Presentation
13th Student Research Conference, Vietnam National University, Hanoi.
Best Presentation Award
5th International Science Forum 2021, HCMC Youth Forum, Vietnam.
Contact
Let’s connect
For research collaboration in autonomous driving, robotics, and physical AI, or for academic opportunities, please reach out by email or through my academic profiles.