newsgathers

A new artificial intelligence framework has been developed by University of Michigan researchers, allowing human-like robots to independently traverse difficult and varied landscapes. This marks a significant advancement in robotic capabilities, moving beyond their previous limitations of flat, predictable surfaces. The innovation holds considerable promise for practical applications, particularly in emergency response and ecological observation.

Historically, robots have struggled with complex outdoor environments, requiring extensive human input for even basic movement. The new AI system, dubbed LEGO-H, integrates the dual challenges of route planning and physical movement into a single learning process. This unified approach enables robots to adapt their gait and direction based on real-time environmental data, a stark contrast to earlier models that separated these functions. This means the robots can autonomously decide whether to walk, hop, step, or jump, reacting dynamically to obstacles like loose rocks, steep inclines, and tight passages.

Kwan-Yee Li, the lead researcher on the study, highlighted that this integrated learning framework allows the robots to develop bespoke movement strategies without any prior human coding. The robots are equipped with visual sensors, GPS for direction and distance, and an innate understanding of their own physical dimensions, balance, and joint movements. This 'body awareness' is a critical component, enabling them to instinctively adjust to challenges. For instance, the robots learned to lean sideways to navigate narrow gaps and could independently recover balance after a stumble.

Co-author Stella Yu emphasized the profound shift this represents, stating that their model provides humanoid robots with the unprecedented ability to perceive, decide, and act entirely on their own. She noted that previous humanoid robots were largely 'blind,' relying heavily on human operators for every decision regarding their movement. In simulated trials, robots of various sizes successfully navigated virtual hiking trails of increasing complexity, demonstrating high levels of completeness, safety, and efficiency. Their performance often equaled or surpassed that of robots provided with pre-existing terrain knowledge, underscoring the importance of intrinsic body awareness.

While current developments have kept the robots' upper bodies relatively fixed to simplify initial modeling, the next phase of research will focus on incorporating full upper-body movement. This enhancement aims to achieve complete full-body coordination, further expanding the robots' versatility. The long-term implications of this technology are substantial, potentially revolutionizing search and rescue efforts in hazardous or disaster-stricken areas. Additionally, these intelligent robots could play a crucial role in monitoring remote and sensitive ecosystems, providing valuable data in areas difficult or dangerous for humans to access.