Intelligent Embodied Agent
Object retrieval in interactive environments with 91% success rate
Intelligent Embodied Agent for Object Retrieval (2022)
Developed an enhanced navigation and manipulation system for object retrieval in interactive environments, achieving high success rates across diverse scenarios.
Key Results
- 91% success rate across 5000 test scenarios
- Handles complex interactions: container opening, tool usage, obstacle navigation
- Spatial reasoning for object localization and tool selection
Technical Contributions
- Spatial reasoning algorithms for efficient object localization
- Tool selection and usage policies for complex retrieval tasks
- Integration of perception, planning, and manipulation
Team Leadership
Led algorithm development while mentoring a team of 5 on implementation and integration, demonstrating both technical depth and collaborative skills.
Environment
Built and tested in AI2Thor interactive simulation environment, handling realistic physics and object interactions.