This work aims to address fundamental questions and create solutions to improve human ability and safety in dangerous unstructured environments. People are inherently cognitively and physically limited. Moreover, we are often perceptually saturated, limiting our ability to respond to dynamic obstacles such as falling debris, runaway vehicles, or intelligent adversaries. This thesis research addresses these mental and physical limitations through three aims. In Aim 1, we investigate how to effectively communicate with people with various perceptual cues to enable more effective evasion behaviors. We then present, optimize, and validate a human-centric motion planner which further improves human ability. In Aim 2, we design machine-learning-based intention recognition algorithms to identify discrete directional motions on offline data and identify lower dimensional motion primitives for real-time control. In Aim 3, we design, characterize, and validate a quasi-direct drive hip exoskeleton on several activities ranging from cyclic to dynamic tasks. Long term, these aims could be integrated into an environmentally aware system of mobile robots monitoring the environment and feeding information to a situation awareness enhancing active exoskeleton that can assist in daily tasks while also protecting operators from workplace to war zone.