Huy photo

Nguyen Dinh Truong Huy
I am currently working as a Research Engineer
at Living Analytics Research Centre, SMU.


Ph.D. Candidate
AI Lab 3 (previously Medical Computing Lab)
School of Computing, National University of Singapore
Email: truong[h][u][y] @ gmail

Thesis work

Supervisors: Prof Leong Tze Yun, Prof Lee Wee Sun

Collaborative Action Planner using Intention Recognition (CAPIR)
We address the Lead-Assistant Collaboration problem that characterizes an autonomous assistant aiding an agent in achieving multiple goals in a known environment; the main source of uncertainty is the agent's drifting intention and behavior model in achieving the goals. We propose the CAPIR framework (Collaborative Action Planner using Intention Recognition) to solve this problem by tracking the assisted agent's intention using inverse planning (a concept inherited from cognitive science), and constructing actions based on the assistant's belief on the agent's intended goal and its knowledge of solving each goal (modeled as a Markov decision process). For complicated goals that yield large planning space, simulation-based algorithms are devised to approximately solve the goals online. Evaluation with human players in an experiment game shows that the assistant's resultant course of actions is near human-level.

Source codes and Executables
The final product is a toolkit, namely CAPIR, used for authoring the AI assistant's behavior.

  1. CAPIR User Manual
  2. CAPIRSolver (not available, pending for release approval): C++ codes used to compute optimal policies of subgoal MDPs. API
  3. Sheep Savior game: Mac OSX, Win32
  4. Unity codes (not available, pending for release approval): Solution files and C# scripts. API

Related links
Demo Clips


Related publications