Nature-inspired navigation system helps robots traverse complex environments without GPS

Robots could soon be able to autonomously complete search and rescue missions, inspections, complex maintenance operations and various other real-world tasks. To do this, however, they should be able to smoothly navigate unknown and complex environments without breaking down or getting stuck, which would require human intervention.

Most autonomous navigation systems rely on global positioning systems (GPS), which can provide information about where a robot is located within a map. In many environments, however, including caves, unstructured spaces and collapsed buildings, GPS systems either do not work or become unreliable.

Researchers at Beijing Institute of Technology recently developed a new nature-inspired system that could improve robot navigation in unstructured and complex environments, without relying on GPS technology. Their proposed framework—outlined in a paper set to be published in Cell Press and currently available on the SSRN preprint server—is inspired by three distinct biological navigation strategies observed in insects, birds and rodents.

“Our research was inspired by a critical gap we identified in the field of bio-inspired robotics,” Sheikder Chandan, first author of the paper, told Tech Xplore. “While many studies have successfully isolated and implemented navigation strategies from individual animals, like an ant’s path integration or a rat’s cognitive mapping, this reductionist approach misses a fundamental biological principle known as ‘degeneracy.’ In nature, robust navigation emerges from the hierarchical integration of multiple, non-identical, yet functionally overlapping strategies.”

A three-part, nature-inspired framework

Instead of developing a system inspired by one navigation strategy observed in a specific category of animals, Chandan and his colleagues wished to create a unified neuromorphic framework that drew from various species. Ultimately, they were able to emulate biological processes that support navigation in insects, birds and rodents.

“We aimed to synthesize the most effective strategies observed in these three categories of animals into a single system, to directly address the core limitations of conventional navigation, such as sensory brittleness and high energy consumption, particularly in challenging, GPS-denied environments,” said Chandan.

The team’s framework thus has three main bio-inspired components that collectively support a robot’s navigation. These are an insect-inspired path integrator, a bird-inspired multisensory fusion system and a rodent-inspired mapping system.

“First, the insect-inspired path integrator, built as a spiking neural network on low-power neuromorphic hardware, acts as a robust internal step-counter for egocentric tracking,” explained Chandan. “The avian-inspired multisensory fusion system then mimics how migratory birds use multiple cues, using a Bayesian filter to dynamically combine inputs from a quantum magnetometer, a polarization compass, and vision, to ensure a reliable heading direction even if one sensor fails.

“Third, a rodent-inspired cognitive mapping system creates a spatial memory by only updating the map upon detecting salient landmarks, mirroring the energy efficiency of the brain’s hippocampus.”

To assess the potential of their nature-inspired framework, the researchers carried out extensive field trials using 23 different robotic platforms. These tests were performed in complex real-world environments, including abandoned mines and dense forests.

“The system was benchmarked against conventional SLAM (Simultaneous Localization and Mapping) and showed a 41% reduction in positional drift, up to 60% higher energy efficiency, and could recover from sensor failures 83% faster,” said Chandan. “Its unique advantage is ‘degeneracy’—when one component is compromised, the others seamlessly take over, providing a level of fault tolerance that isolated systems lack.”

Performance gains and possible applications

In initial field tests, the architecture developed by this team of researchers was found to achieve remarkable results, allowing a wide range of robots to successfully navigate unstructured and difficult environments.

“We didn’t just improve a single algorithm; we created a new systems-level paradigm that is inherently more resilient,” said Chandan. “Quantitatively, this resulted in significant, simultaneous gains in accuracy, energy efficiency, and robustness across diverse robotic platforms. A key demonstration was the system’s rapid recovery from sensor failure; when the primary camera was blinded, it re-established accurate positioning in just over 3 seconds by leveraging its other functional subsystems.”

In the future, the framework developed by Chandan and his colleagues could be improved further and deployed on an even larger pool of robotic systems, allowing them to reliably and autonomously tackle missions in unpredictable environments. In addition, it could inspire the creation of similar robot navigation systems that draw from the navigation strategies employed by a variety of animals.

“This work provides a formal blueprint for creating machines with true ‘ecological fluency,’ capable of long-term operation in environments where failure is not an option,” said Chandan. “This could include applications in disaster response, such as navigating collapsed buildings, planetary exploration on other worlds, and deep-sea missions, where conventional GPS and perfect sensing are unavailable.”

The researchers are currently planning new studies aimed at overcoming some observed limitations of their framework. For instance, they would like to integrate on-chip and continuous learning to make the navigation of robots even more lifelike and adaptable.

“Currently, our system’s neural weights are largely pre-configured, but biological systems continuously learn and adapt through synaptic plasticity,” added Chandan. “We plan to explore emerging technologies like memristive synapses to incorporate this capability directly into the hardware.

“Furthermore, we aim to scale the system for kilometer-scale environments, which will require developing more sophisticated memory organization schemes to handle larger spatial maps efficiently. Our ultimate goal is to create robots that don’t just mimic isolated animal behaviors but embody the continuous learning and scalability of biological intelligence.”

© 2025 Science X Network