Ultra-wideband (UWB) technology is increasingly used to build location-aware IoT applications because of its outstanding positioning accuracy. Its communication performance, however, is unexplored and strongly affected by the chosen physical layer settings as well as by the surrounding environment. Finding an effective way to increase the dependability of UWB communications is yet an open problem. In this paper, we study the performance of different UWB physical layer settings and use them as tuning knobs to increase the energy efficiency and robustness of communications. Towards this goal, we first experimentally quantify the reliability and energy cost of each setting, in order to understand which physical layer configuration to privilege depending on the application requirements. We then use the estimated channel impulse response-a unique feature of UWB transceivers-to accurately measure the link quality and to extract relevant information about the characteristics of the surrounding environment, such as the presence of destructive interference. Capitalizing on this information, we design a scheme that adapts the UWB physical layer settings at runtime. An experimental evaluation using the Decawave DW1000 radio shows the effectiveness of the proposed adaptive scheme, highlighting the increased communication robustness and energy efficiency.