Recruiting Heterogeneous Crowdsource Vehicles for Updating a High-definition Map
A novel 'threshold-type' policy for recruiting cars reduces average costs by 19% and speeds up computation by 14%.
A research team from The Chinese University of Hong Kong, Shenzhen, and the University of Macau has published a paper presenting a novel, cost-optimized framework for maintaining the high-definition (HD) maps essential for autonomous driving. Instead of relying on expensive, dedicated mapping fleets, the system crowdsources data from everyday vehicles, which vary in sensor quality and operational cost. The core challenge is balancing map 'freshness' (how up-to-date it is) with the cost of recruiting these heterogeneous vehicles. The researchers solved this by modeling the problem as a Markov decision process (MDP).
Their analysis yielded a counter-intuitive yet optimal 'threshold-type' policy, where a company should sometimes recruit vehicles earlier if they arrive more frequently or have higher costs or better sensors. To compute this policy efficiently, they developed the Bound-based Relative Value Iteration (BRVI) algorithm. In simulations, this optimal policy achieved a 19.04% reduction in average cost compared to existing methods. Furthermore, the BRVI algorithm itself converged to the solution 13.66% faster on average than previous computational approaches, making real-time deployment more feasible.
- The optimal recruitment policy for heterogeneous vehicles reduces average HD map update costs by 19.04% versus state-of-the-art methods.
- The novel Bound-based Relative Value Iteration (BRVI) algorithm finds this optimal policy 13.66% faster than existing computational solutions.
- The research reveals counter-intuitive strategies, like recruiting earlier when vehicle costs are higher, to optimize the freshness-cost tradeoff.
Why It Matters
This makes scalable, real-time HD maps for autonomous vehicles significantly cheaper and faster to maintain, accelerating commercial deployment.