Project

We are building the planning library for the Autolife robot, using OMPL as the frontend and VAMP as the backend. The goal is millisecond-level CPU motion planning with support for ASAP planning, subgroup planning, whole-body planning, and richer constraint handling, aiming to push past current GPU-heavy baselines such as cuRobo.

TyGrit is a unified playground for mobile manipulation. We are implementing model-based, model-free, reinforcement learning, and imitation learning baselines within a single framework, and will generate and release accompanying datasets. Everything in TyGrit is designed end-to-end for mobile manipulation (MoMa).

We are exploring skill representations that enable data-efficient and general policies. We represent tasks via the objects themselves, decoupled from robot and camera viewpoints, making the policy agnostic to view changes, appearance/illumination changes, embodiment changes, etc.

This project combines SIMD motion planning with hierarchical multilayer RRTC to enable the Fetch robot to perform whole body planning in under 100ms. This breakthrough makes near real-time motion planning possible for more reactive mobile manipulation tasks, significantly enhancing the robot's responsiveness in dynamic environments.

A comprehensive framework for robot imitation learning that supports multiple data gathering interfaces (GELLO, VR controller, joint stick, and hand tracking) and various robot embodiments (Fetch, Kinova Gen3, Franka FR3). The behavior cloning component implements state-of-the-art approaches including ACT and diffusion policy, enabling efficient transfer of human demonstrations to robotic systems.

A comprehensive digital twin platform designed to support student projects in CS6244 Advanced Topics in Robotics at NUS. This platform provides the infrastructure and tools for graduate students to develop and evaluate mobile manipulation algorithms in realistic environments, serving as the foundation for various course projects and research explorations.

An enhanced Gaussian Splatting framework for indoor 3D mesh reconstruction, incorporating novel geometric constraints for improved accuracy. The toolkit provides comprehensive documentation and easy-to-use interfaces for various 3D reconstruction tasks.

Research on improving Visual Place Recognition (VPR) systems through enhanced local descriptor selection. Developed novel descriptor selection methods using semantic segmentation and high-pass filters, creating a comprehensive evaluation framework for VPR performance.