Flag indicating the english language is selected

Aryan Mangal

Hello! I am a Machine Learning Engineer at Woven by Toyota in Palo Alto. I currently work on foundation models for driving behavior, with a focus on pretraining and large-scale training and inference optimization. I recently completed my MS in Robotics at The Robotics Institute, Carnegie Mellon University (GPA 4.08/4), advised by Prof. Katia Sycara. Previously, I worked on human behavior anticipation for effective human–robot collaboration.

I received my B.Tech in Electronics and Communication from Delhi Technological University (GPA 9.5/10). As an undergrad I led the Unmanned Ground Vehicle team at DTU and was a 2022 CMU Robotics Institute Summer Scholar with Prof. Sebastian Scherer and Prof. Chen Wang.

Palo Alto, CA, USA

Email / LinkedIn / Google Scholar / GitHub / CV

Aryan Mangal, machine learning engineer at Woven by Toyota

News

Aug 2026 Competed at Grokathon (xAI, San Francisco), building grokbot — personalized prediction-market insights from X activity using Grok. Code
Jul 2026 Took 2nd place at the Embodied Metal robotics hackathon in San Francisco, building an end-to-end soccer goalkeeper on a Unitree G1. Demo
Apr 2026 Competed at RoboHacks at Y Combinator, building Rocky, a voice-driven mobile manipulation system on Innate-OS. Demo
Apr 2026 Served as a reviewer for IROS 2026.
Nov 2025 Served as a reviewer for ICRA 2026.
Jul 2025 Joined Woven by Toyota as a Machine Learning Engineer on the Foundational Behavior Model.
May 2025 Graduated with an MS in Robotics from CMU (GPA 4.08/4), advised by Prof. Katia Sycara.
Aug 2023 Started MS in Robotics at Carnegie Mellon University.
Mar 2023 JN Tata Endowment for Higher Education — Top 0.1% in a pool of 10,000+ applicants.
Mar 2022 Robotics Institute Summer Scholarship — One of 4 students from India selected for the CMU RISS research internship.
Mar 2022 Intelligent Ground Vehicle Competition — 1st place in the Cybersecurity Challenge at IGVC, Oakland University.
Mar 2022 DAAD WISE Internship — One of 80 students from India selected for research in Germany.
Mar 2022 Mitacs Globalink Internship — One of 117 students from India selected for an internship in Canada.
Oct 2021 MATLAB Minidrone Competition — International Finale qualifier; top 8 from Europe, Middle East, and Asia.

Experience

Industry

Machine Learning Engineer — Foundational Behavior Model Jul 2025 – Present

Woven by Toyota · Palo Alto, CA

  • Built FBM (Foundational Behavior Model), a foundation model for driving behavior powering multiple downstream tasks via task-specific heads: distillation to the onboard behavior model, synthetic scene generation, and scene search.
  • Built and deployed a neural simulation model trained with offline RL to enable inference-time control of agent behavior for closed-loop evaluation, and generated 2,000+ hours of synthetic scenario data.
  • Accelerated large-scale Transformer training by migrating the compilation stack from TorchScript to TorchDynamo (torch.compile) and integrating FlashAttention-2 for memory-efficient attention.
  • Built an agentic library that automates org-wide ML workflows (data generation, on-vehicle deployment validation, model training), cutting the ML model lifecycle from weeks to days and adopted by 700+ employees.
  • Led integration of navigation intent into the ML behavior model, enabling action decisions without rule-based heuristics.
Computer Vision Intern — 3D Bounding-Box Detection for Unseen Objects Jan 2023 – May 2023

Orangewood Labs · New Delhi, India

  • Engineered a solution to generate safe trajectories for robots to interact with novel objects, using RGB-D images.
  • Leveraged Segment Anything Model’s 2D bounding boxes and depth data to generate 3D boxes for objects.
Robotics Intern — UAV Localization Apr 2021 – Jul 2021

Octobotics Tech Private Limited

  • Built UAV localization for dim-light and GPS-denied environments in ROS.
Research Intern — EV Motor Controllers Jan 2021 – Jul 2021

Ministry of Electronics and Information Technology

  • Designed motor controllers for low- and medium-power EVs; heat-sink work cut MOSFET average temperature by 10°C.

Research

Dual-arm setup for dexterous VLA with closed-loop VLM control
Zero-Human Supervision: Closed-Loop VLM Control for Dexterous VLA Policies 2026 – Present
  • Closed-loop VLM control and recovery for dexterous VLA, with no human demonstration or intervention.
SymAnt: Neuro-Symbolic Diffusion Models for Action Anticipation Aug 2023 – Jul 2025

Advanced Agents Robotics Technology Lab, CMU · code · paper

  • Built diffusion models to anticipate humans’ long-term actions from short video, enabling proactive robot assistance.
  • Achieved SOTA accuracy (+4–5%) at 80× smaller size and 3× faster training, enabling real-time robot deployment.
PyPose: A Library for Deep Learning with Physics-based Optimization Jan 2022 – Dec 2022

The AirLab, CMU, Pittsburgh · code · CVPR paper

  • Developed a PyTorch library exposing physics-based optimization and Lie-group operations as differentiable layers.
  • Built a batched differentiable IMU preintegration layer (SO(3) propagation, Jacobian covariance) for neural training.
AirLoc: Object-based Indoor Relocalization May 2022 – Present

Robotics Institute Summer Scholar · The AirLab, CMU · Prof. Sebastian Scherer · advisor

  • Developed AirLoc, indoor relocalization using geometric and semantic object cues; outperformed SOTA in room-level relocalization.
Qualitative flare-removal results from the paper
Self-Supervised Flare Removal 2022 – 2023

Delhi Technological University · code · paper

  • Flare removal with conditional GANs and self-supervised fine-tuning; SSIM improved from 0.887 to 0.916.
Place Recognition with Scene Graphs May 2021 – Dec 2021

IIIT Hyderabad · Dr. K. Madhava Krishna · advisor

  • Place recognition with scene graphs and deep learning for room- and metric-level localization. Baselines: NetVlad, Scan Matching, DeLG.
Mental Health Recognition using HRV and GSR May 2021 – Present

Samsung Digital Academy Research Lab, DTU · Dr. Divyashikha Sethia · advisor

  • Developed a stress-level classification system from ECG-derived HRV and GSR features, reaching 93.96% binary and 81.41% three-class accuracy.

Projects

Soccerbot Jul 2026

Embodied Metal hackathon, 2nd place · Unitree G1 · code · website · demo

  • End-to-end goalkeeper on a Unitree G1: learned 14-DOF ACT ball pickup, then a scripted turn, avoid, and throw.
  • Depth-based human detection with side-to-side shuffle until the throwing lane is clear.
Rocky Apr 2026

Voice-driven manipulation and visual search · code · demo

  • Trained manipulation policies (ACT, GROOT) for can pickup robust to arbitrary orientations.
  • End-to-end voice-to-action for multi-step pickup and visual search with spatial memory.
Xiao Hei VLN: RViz, Unity sim, and exploration logs
Xiao Hei 2026

CMU Vision-Language Navigation Challenge · code · website

  • VLN stack with a Gemini responder, YOLO-World + SAM perception, a 3D object scene graph, and frontier-based exploration.
Sparse-dense matrix multiply diagram for Needle CUDA kernels
Needle 2024

Deep Learning Systems · code

  • Autograd library with custom CUDA kernels for sparse-dense matrix multiplications; 40% faster GNN training vs dense matmul.
Legal document summarization pipeline: LoRA fine-tune then FAISS RAG
Legal Document Summarization 2024

Carnegie Mellon University · code

  • Fine-tuned Llama-2-7b with LoRA and deployed a FAISS/TF-IDF RAG pipeline for Indian legal documents.
Griffin 2019 – 2023

Team Leader · DTU Unmanned Ground Vehicle · code

  • ROS navigation and perception with SLAM, GPS-waypoint navigation, and LiDAR–IMU–GPS fusion.
  • Deep networks for lane following and obstacle avoidance; 3rd at IGVC 2023.
Grokbot Aug 2026

Grokathon · xAI · code

  • Polymarket × X opinion pipeline: harvest discussion, Grok camp reports, and an X-style Markets UI.
  • Auto explainer videos from camp reports with Grok Imagine and ffmpeg.
Parrot minidrone line-follower simulation
Line Follower Minidrone Apr 2021 – Jun 2021

MATLAB Minidrone Competition · code

  • Model-based line-follower algorithm in MATLAB/Simulink for a Parrot minidrone.
FPGA development board
Neural Networks on FPGAs Jan 2021 – May 2021

DTU

  • Compared neural network implementation methods on FPGAs for early prototyping in Python.
Micromouse maze in Gazebo
Micro-Mouse Challenge Dec 2020

IIT Bombay TechFest · code

  • Maze exploration in Gazebo with backtracking and Slam_gmapping in ROS.

Publications

2025

Parameter-Efficient Neuro-Symbolic Action Anticipation via Iterative Context Refinement

Aryan

Master's thesis, Carnegie Mellon University, CMU-RI-TR-25-45 · pdf

2025

SymAnt: Neuro-Symbolic Diffusion Models for Action Anticipation

Aryan, Simon Stepputtis, Yaqi Xie, Hossein Nourkhiz Mahjoub, Katia Sycara

Workshop — ICRA 2025 · Under review, WACV 2027 · paper · code

2024

Symbolic Graph Inference for Compound Scene Understanding

Aryan, Simon Stepputtis, Sarthak Bhagat, Joseph Campbell, Kwonjoon Lee, Hossein Nourkhiz Mahjoub, Katia Sycara

arXiv 2410.22626 · Workshop — ICRA 2024 · paper · code

2023

PyPose: A Library for Robot Learning with Physics-based Optimization

Chen Wang, Dasong Gao, Kuan Xu, Junyi Geng, Yaoyu Hu, Yuheng Qiu, Bowen Li, Fan Yang, Brady Moon, Abhinav Pandey, Aryan, Jiahe Xu, Tianhao Wu, Haonan He, Daning Huang, Zhongqiang Ren, Shibo Zhao, Taimeng Fu, Pranay Reddy, Xiao Lin, Wenshan Wang, Jingnan Shi, Rajat Talak, Kun Cao, Yi Du, Han Wang, Huai Yu, Shanzhao Wang, Siyu Chen, Ananth Kashyap, Rohan Bandaru, Karthik Dantu, Jiajun Wu, Lihua Xie, Luca Carlone, Marco Hutter, Sebastian Scherer

CVPR 2023 · paper · code

2023

Thermal Study of MOSFET Heat Sinks for Motor Controller in Neighbourhood Electric Vehicles

Kshitij Tripathi, Aakash Ghosh, Aryan Mangal, Mini Sreejeth, S. Indu

International Journal of Electric and Hybrid Vehicles 15 (3), 256–271 · paper

2023

Image Flare Removal using Deep Convolutional Generative Adversarial Networks

Aryan, Ishan Prakash, S. Indu, Mini Sreejeth

IEEE SPIN 2023 · paper · code

2023

AirLoc: Object-based Indoor Relocalization

Aryan, Bowen Li, Sebastian Scherer, Yun-Jou Lin, Chen Wang

arXiv 2304.00954 · paper · code

2022

Grad-Lidar-SLAM: Fully Differentiable Global SLAM for Lidar

Aryan, Dheeraj Vattikonda, Erqun Dong, Sabyasachi Sahoo

IROS 2022 Workshop on Probabilistic Robotics in the Age of Deep Learning · paper · code

2022

HRV and GSR as Viable Physiological Markers for Mental Health Recognition

Aryan, Lovish Arya, Divyashikha Sethia

COMSNETS 2022 · paper

2021

Realization of Low-Cost Footwear Integrated Step-Counting Device for Health Monitoring System

Aryan, Poornima Mittal

Journal of Physics: Conference Series 1998 (1), 012006 · paper