https://zhaofanbo.com/en/projects/Recent content in Selected projects on Fanbo ZhaoHugoenSun, 30 Aug 2026 00:00:00 +0000https://zhaofanbo.com/en/projects/smartfly/Sun, 30 Aug 2026 00:00:00 +0000https://zhaofanbo.com/en/projects/smartfly/<p><strong>Research in progress · August 2026 snapshot</strong></p> <p>SmartFly asks: <strong>how can an agent use vision, odor, and wind to stay stable and find a target?</strong> Built around a MuJoCo fly simulation, the project combines a learned high-level navigation policy with a frozen low-level flight controller to explore the relationship between perception, memory, and action.</p> <h2 id="research-questions">Research questions</h2> <ul> <li><strong>Can one policy support two navigation tasks?</strong> The visual task provides relative target information. The vision–odor task hides the source location and uses odor, apparent wind, and vision for source seeking and obstacle avoidance. The goal is to support both tasks with the same checkpoint.</li> <li><strong>How do visual priors relate to representations and behavior?</strong> The comparison uses a trainable CNN and FlyVis with a frozen pretrained backbone, under matched tasks and training budgets.</li> <li><strong>What separates a demonstration from reliable evidence?</strong> Stable-flight replays help inspect closed-loop behavior and actual policy inputs. Full navigation claims require independent testing and statistics across training seeds.</li> </ul> <h2 id="current-implementation">Current implementation</h2> <p>The project uses <strong>Python, PyTorch, Ray RLlib / PPO, MuJoCo, and dm_control</strong>, with four main components:</p>