Exoskeletons could make walking easier for people including military personnel, first responders, older adults, and people with disabilities. These devices have shown some improvements by reducing the energetic cost of walking, but we still don’t know what assistance strategies are best or how effective they can be. Which leg joints should exoskeletons assist? What torques should they apply? What’s the greatest improvement we could expect? In what conditions can they be most effective? To investigate these questions, we designed a bilateral lower-limb exoskeleton emulator to study walking assistance. We then optimized hip-knee-ankle exoskeleton assistance for each joint individually, for two-joint combinations, and for the whole-leg. We found that assisting the whole leg reduced the energy cost of walking by 50%, double the state-of-the-art. However, while assisting the whole-leg was most effective, we found assisting a single joint to be more efficient in terms of percent reduction per joint assisted. To study how this optimal assistance could change in new environments, we then optimized assistance for walking on inclines up to 15 degrees. We found similar improvements of at least 50%, demonstrating whole-leg assistance can be beneficial across a wide range of inclines. The optimized assistance strategies we found could be used to inform the design of future exoskeletons, bringing them closer to the goal of product-like devices for those who would benefit from walking assistance.