Modern navigation algorithms based on deep reinforcement learning (RL) show\npromising efficiency and robustness. However, most deep RL algorithms operate\nin a risk-neutral manner, making no special attempt to shield users from\nrelatively rare but serious outcomes, even if such shielding might cause little\nloss of performance. Furthermore, such algorithms typically make no provisions\nto ensure safety in the presence of inaccuracies in the models on which they\nwere trained, beyond adding a cost-of-collision and some domain randomization\nwhile training, in spite of the formidable complexity of the environments in\nwhich they operate. In this paper, we present a novel distributional RL\nalgorithm that not only learns an uncertainty-aware policy, but can also change\nits risk measure without expensive fine-tuning or retraining. Our method shows\nsuperior performance and safety over baselines in partially-observed navigation\ntasks. We also demonstrate that agents trained using our method can adapt their\npolicies to a wide range of risk measures at run-time.\n