Project 1: Vision-based Planetary Path Planning Algorithm
- Belongs to “Research on Navigation and Guidance Control of Precise Landing on Planetary Surface”, 973 National Basic Research Program of China
- Worked as the main researcher of this sub-project
- Designed a novel deep neural network architecture with double branches to plan path for planetary rovers directly from orbital images
- Demonstrated that the proposed deep neural network architecture achieved better performance and faster convergence than the existing ones and generalized well to unknown environment.
- Included the research results in Prof. Yuanqing Xia’s monograph, and responsible for the compilation of related contents.
Project2: Learning-based Mobile Crowdsensing Games
- Belongs to key project in Beijing and international cooperation project
- Cooperated with Dr. Yufeng Zhan
- Designed the pricing and sensing time allocation strategies for MCS (mobile crowdsensing) systems with multiple TIs and multiple MUs (mobile users) to incentivize MUs for participation, and studied this problem from a free market perspective with the goal of achieving a SE (Stackelberg Equilibrium).
- Formulated the incentive mechanism as a Stackelberg game, which can reveal the characteristics of the supply-demand pattern for MCS.
- Analyzed and proved that there exists a SE with a lack of closed-form expression, and proposed a DRL (deep reinforcement learning) based solution called DDIM (DRL based Dynamic Incentive Mechanism) to solve it.
- The extensive simulation results showed that DDIM outperforms the state-of-art and baseline approaches.
Project 3: Modelling Traffic Flow with Intelligent Vehicles
- Belongs to Problem C of MCM’2017
- Meritorious Winner (Top 9%, global)
- Worked as team leader
- Established the Modern Dynamic Model of Traffic Flow (MDTF) and Smart Traffic Flow Model Based on Cellular Automata (STCA) to analyze the traffic flow mixed with self-driving cars (SDV) and non-self-driving cars (NSDV).
- Introduced a new variable (the proportion of SDVs), analyzed two typical situations (the diffusion of traffic jam and the effects of ramps), and concluded that SDVs could lighten the diffusion of traffic jam and smooth the density distribution of traffic flow.
- Divided the information a vehicle may receive into two types (in-horizon information and out-horizon information), abstracted two laws (moving and changing lanes) to depict the mechanism of traffic flow mixed with SDVs, defined a synchronization effect between two SDVs, proposed four indexes to evaluate the effects of SDVs on traffic flow, and fitted the function between each index and the proportion of SDVs.
- Combined location data in Excel spreadsheet with speed data in Washington State Speed Report, compared the simulation results with real data and predicted the effect of SDVs.