Multi-robot Task Allocation (MRTA) is the problem of assigning tasks to robots subject to a performance objective. Among existing approaches to MRTA, auction-based methods are widely used. In an auction-based method, each robot typically computes its Euclidean distance to all the given tasks, and those values are bids based on which a global auctioneer allocates the tasks to them. Although simple to compute, these approaches result in an inefficient navigation of robots to reach the tasks in an environment populated with humans. We overcome this limitation by augmenting bids in an auction-based MRTA method with knowledge of human motions. As a result, this augmented task allocation method may, for instance, assign a task to a robot which is further away so long as the robot avoids possibly congested places. We validate the approach through simulated fleets of robots in a shopping centre and a small-scale warehouse environment. Our results show significant improvement over the allocation that ignores knowledge of human dynamics.