I'm a final-year PhD student at MIT CSAIL, advised by Leslie Kaelbling and Tomás Lozano-Pérez. My research focuses on planning and perception for robotics, aiming to enable robots to reason about long-horizon tasks and generalize effectively in the real-world.
Previously, I was a research intern at NVIDIA and Apple, and a software engineer at Amazon Web Services. I completed my undergraduate degree at the Australian National University, where I worked on learning for planning and search. I'm originally from New Zealand 🇳🇿.
You can contact me on my email willshen@mit.edu.
CoRL 2026 Agentic Robotics Workshop
Feijoa is an embodied language agent that grounds objects at test time by querying and refining over a posed RGB-D scan, then executes navigation and manipulation skills through a harness that checks feasibility and verifies effects.
arXiv preprint, 2026
TiPToP is a modular, open-vocabulary system that combines vision foundation models with Task and Motion Planning to solve long-horizon manipulation tasks from RGB images and natural-language instructions. It requires zero robot data, matches or outperforms a state-of-the-art VLA, and is easily adapted to new embodiments.
IEEE Robotics and Automation Letters (RA-L), 2026
We leverage the common-sense reasoning and code-generation abilities of VLMs by using them to write constraints that plug into TAMP solvers at runtime. This enables us to tackle 'open-world' tasks beyond TAMP's predefined concepts.
Robotics: Science and Systems (RSS), 2025
We leverage parallelism to simultaneously explore thousands of TAMP solutions using gradient information. This enables our algorithm to solve difficult, tightly constrained long-horizon planning problems.
Also presented at the CoRL 2024 Workshop on Differentiable Optimization Everywhere.
Conference on Robot Learning (CoRL), 2023Best Paper Award
We distill features from 2D foundation models into 3D feature fields, and enable few-shot language-guided manipulation that generalizes across object poses, shapes, appearances and categories.
Also presented at the CVPR 2023 Workshop on 3D Vision and Robotics.
International Conference on Automated Planning and Scheduling (ICAPS), 2020
We present STRIPS-HGN, the first approach capable of learning domain-independent planning heuristics from scratch using hypergraph neural networks for search.
Also presented at the AAAI GenPlan20 workshop.
Symposium on Combinatorial Search (SoCS), 2019
We combine generalized neural network policies (Action Schema Networks) with search algorithms (MCTS) to exploit the strengths and overcome the weaknesses of each to solve probabilistic planning problems.
Also presented at the ICAPS HSDIP'19 workshop.
Outside of research, I enjoy skiing ⛷️, hiking 🥾, and playing tennis 🎾 and badminton 🏸.