Kaito Namatame - Portfolio

Hi, I'm Kaito 👋

I'm currently a Master's student in Robotics and Autonomous Systems at ASU (Arizona State University). I enjoy working across hardware and software, and I like learning by building and experimenting. Outside of robotics, I enjoy running, playing soccer or pickleball, or watching Formula 1 races 🏎️

Education Projects Skills Contact

Education

Arizona State University

Apr. 2024 - May. 2026

M.S. in Robotics & Autonomous Systems

Research Theme: LLM Generated Parametric Bias for VLA

GPA: 3.85 / 4.0

Takushoku University

Apr. 2020 - Mar. 2024

B.S. in Mechanical Engineering

Research Theme: Development of Ant-like Hexapod Robot

GPA: 3.29 / 4.0

Projects

Development of Ant-like Hexapod Robot

Code

Designed and built a six-legged walking robot inspired by ant locomotion, with 3 DOF per leg.
Developed control algorithms utilizing both forward kinematics and inverse kinematics for leg movements.
Experimented with various gait patterns and analyzed performance in terms of speed, distance, and stability.

Robotics Kinematics Embedded C++ CAD

Semantic 3D Point Cloud Mapping

Code

Developed a semantic 3D point cloud by integrating 3D point clouds from ORB-SLAM3 with semantic segmentation masks.
Labeled each point with corresponding semantic categories (U-Net + VGG16), enhancing geometric information.
Demonstrated improved performance compared to using semantic segmentation or ORB-SLAM3 alone.

Python PyTorch ORB-SLAM3 U-Net 3D Perception

LLM-Guided Modulation for Failure Recovery in VLA

Ongoing Research

Proposed a failure recovery framework for Vision-Language-Action models by modulating diffusion-based action using FiLM.
Fed previous actions and their corresponding rewards back to an LLM to analyze failure contexts and generate corrective FiLM coefficients.
Enabled adaptive correction of action distributions, transforming failed executions into successful behaviors.

Python PyTorch Diffusion Models FiLM Vision-Language-Action LLM Mujoco

Autonomous Drone for Geological Mapping and Landing

Code

Developed a ROS2 node for autonomous survey using PX4 SITL and onboard RGB-D input.
Performed real-time environment mapping via ORB-SLAM and logged 3D terrain data for post-analysis.
Detected geological features and optimized landing maneuvers.

ROS2 PX4 Python Mapping

Cart Pole Swing-Up Reinforcement Learning

Code

Developed a reinforcement learning controller for the Cart-Pole swing-up task, achieving upright balance while keeping the cart near the center.
Used Proximal Policy Optimization (PPO) to control a nonlinear, underactuated system with continuous actions.
Demonstrated that reinforcement learning effectively learns both swing-up and balance control behaviors.

Genesis ROS2 Python Reinforcement Learning

Grid World Q-Learning

Code

Implemented a Q-learning–based agent for the 3×4 Gridworld problem to learn an optimal policy.
Modeled a stochastic environment with probabilistic action outcomes, including goal, losing, and wall states.
Tracked training performance and learning dynamics using Weights & Biases (W&B).

Q-Learning W&B Python Reinforcement Learning

Coordinated Frontier-Based Multi-Robot

Code

Implemented a coordinated frontier-based exploration strategy for multi-robot systems in MATLAB using occupancy grid mapping and lidar sensing.
Designed a shared exploration architecture with frontier detection, distributed goal selection, A* path planning, and conflict-free frontier reservation.
Analyzed performance across 2, 3, and 4-robot teams through simulations.

MATLAB Mulit Robotics Path Planning Frontier-Based Exploration