Self-maintenance The first requirement for complete physical autonomy is the ability for a robot to take care of itself. Many of the battery-powered robots on the market today can find and connect to a charging station, and some toys like Sony's
Aibo are capable of self-docking to charge their batteries. Self-maintenance is based on "
proprioception", or sensing one's own internal status. In the battery charging example, the robot can tell proprioceptively that its batteries are low, and it then seeks the charger. Another common proprioceptive sensor is for heat monitoring. Increased proprioception will be required for robots to work autonomously near people and in harsh environments. Common proprioceptive sensors include thermal, optical, and haptic sensing, as well as the
Hall effect (electric). sensors without human intervention to keep themselves safe and operating properly.
Sensing the environment Exteroception is
sensing things about the environment. Autonomous robots must have a range of environmental sensors to perform their task and stay out of trouble. The autonomous robot can recognize sensor failures and minimize the impact on the performance caused by failures. • Common exteroceptive sensors include the
electromagnetic spectrum, sound, touch, chemical (smell, odor), temperature, range to various objects, and altitude. Some robotic lawn mowers will adapt their programming by detecting the speed in which grass grows as needed to maintain a perfectly cut lawn, and some vacuum cleaning robots have dirt detectors that sense how much dirt is being picked up and use this information to tell them to stay in one area longer.
Task performance The next step in autonomous behavior is to actually perform a physical task. A new area showing commercial promise is domestic robots, with a flood of small vacuuming robots beginning with
iRobot and
Electrolux in 2002. While the level of intelligence is not high in these systems, they navigate over wide areas and pilot in tight situations around homes using contact and non-contact sensors. Both of these robots use proprietary algorithms to increase coverage over simple random bounce. The next level of autonomous task performance requires a robot to perform conditional tasks. For instance, security robots can be programmed to detect intruders and respond in a particular way depending upon where the intruder is. For example,
Amazon launched its Astro for home monitoring, security and eldercare in September 2021.
Autonomous navigation Indoor navigation For a robot to associate behaviors with a place (
localization) requires it to know where it is and to be able to navigate point-to-point. Such navigation began with wire-guidance in the 1970s and progressed in the early 2000s to beacon-based
triangulation. Current commercial robots autonomously navigate based on sensing natural features. The first commercial robots to achieve this were Pyxus' HelpMate hospital robot and the CyberMotion guard robot, both designed by robotics pioneers in the 1980s. These robots originally used manually created
CAD floor plans, sonar sensing and wall-following variations to navigate buildings. The next generation, such as MobileRobots' PatrolBot and autonomous wheelchair, both introduced in 2004, have the ability to create their own laser-based
maps of a building and to navigate open areas as well as corridors. Their control system changes its path on the fly if something blocks the way. At first, autonomous navigation was based on planar sensors, such as laser range-finders, that can only sense at one level. The most advanced systems now fuse information from various sensors for both localization (position) and navigation. Systems such as Motivity can rely on different sensors in various areas, depending upon which provides the most reliable data at the time, and can re-map a building autonomously. Rather than climb stairs, which requires highly specialized hardware, most indoor robots navigate handicapped-accessible areas, controlling elevators and electronic doors. With such electronic access-control interfaces, robots can now freely navigate indoors. Autonomously climbing stairs and opening doors manually are topics of research at the current time. As these indoor techniques continue to develop, vacuuming robots will gain the ability to clean a specific user-specified room or a whole floor. Security robots will be able to cooperatively surround intruders and cut off exits. These advances also bring concomitant protections: robots' internal maps typically permit "forbidden areas" to be defined to prevent robots from autonomously entering certain regions.
Outdoor navigation Outdoor autonomy is most easily achieved in the air, since obstacles are rare.
Cruise missiles are rather dangerous highly autonomous robots. Pilotless drone aircraft are increasingly used for reconnaissance. Some of these
unmanned aerial vehicles (UAVs) are capable of flying their entire mission without any human interaction at all except possibly for the landing where a person intervenes using radio remote control. Some drones are capable of safe, automatic landings, however.
SpaceX operates a number of
autonomous spaceport drone ships, used to safely land and recover
Falcon 9 rockets at sea. Few countries like India started working on robotic deliveries of food and other articles by
drone. Outdoor autonomy is the most difficult for ground vehicles, due to: • Three-dimensional terrain • Great disparities in surface density • Weather exigencies • Instability of the sensed environment
Open problems in autonomous robotics Several open problems in autonomous robotics are special to the field rather than being a part of the general pursuit of AI. According to George A. Bekey's
Autonomous Robots: From Biological Inspiration to Implementation and Control, problems include things such as making sure the robot is able to function correctly and not run into obstacles autonomously. Reinforcement learning has been used to control and plan the navigation of autonomous robots, specifically when a group of them operates in collaboration with each other. ;Energy autonomy and foraging Researchers concerned with creating true
artificial life are concerned not only with intelligent control, but further with the capacity of the robot to find its own resources through
foraging (looking for food, which includes both energy and spare parts). This is related to
autonomous foraging, a concern within the sciences of
behavioral ecology,
social anthropology, and
human behavioral ecology; as well as
robotics,
artificial intelligence, and
artificial life.
Systemic robustness and real-world brittleness Autonomous robots remain highly vulnerable to unexpected changes in real-world environments. Even minor variations like a sudden beam of sunlight disrupting vision systems or unanticipated terrain irregularities can cause entire systems to fail. This brittleness stems from robotics being an inherently
systems problem, where a deficiency in any module (perception, planning, actuation) can compromise the whole robot.
Open-world scene understanding Robots often depend on datasets captured under controlled conditions, limiting their ability to generalize to novel, dynamic real-world scenarios. They struggle with unknown objects, occlusions, varying object scales, and rapidly changing environments. Developing
self-supervised, lifelong learning systems that adapt to
open-world conditions remains a pressing challenge.
Multi-robot coordination and decentralization Scaling robot systems raises thorny issues in coordination, safety, and communication. In
multi-agent navigation, challenges like deadlocks, selfish behaviors, and sample inefficiencies emerge. Innovations such as dividing planning into sub-problems, combining RL with imitation learning, hybrid centralized-decentralized approaches (e.g., prioritized communication learning), attention mechanisms, and graph transformers have shown promise, but large-scale, stable, real-time coordination remains an open frontier.
Simulation-to-real (“reality gap”) transfer Deep reinforcement learning is a powerful tool for teaching robots navigation and control, but training in simulation introduces discrepancies when deployed in reality. The
reality gap (or differences between simulated and real environments) continues to impede reliable deployment, despite strategies to mitigate it.
Hardware and bio hybrid constraints Physical limitations of batteries, motors, sensors, and actuators constrain robot autonomy, endurance, and adaptability, especially for humanoid or soft-bio hybrid robots. While
Biohybrid system (e.g., using living muscle tissue) hint at leveraging biological energy and actuation, they introduce radically new challenges in materials, integration, and control.
Ethics, liability, and societal integration As robots become more autonomous, especially in public or collaborative roles, ethical and legal issues grow. Who is responsible when an autonomous system causes harm? Regulatory frameworks are still evolving to address liability, transparency, bias, and safety in systems like
Self-driving car or socially interactive robots.
Embodied AI and industrial adoption While AI algorithms have made strides, embedding them into robots (embodied AI) for real-world use remains slow-moving. Hardware constraints, economic viability, and infrastructure limitations limit widespread adoption. For instance, humanoid robots like “
Pepper (robot)” failed to achieve ubiquity due to fundamental cost and complexity issues.
Human-robot interaction in unstructured settings Robots working near or alongside people in a shared space need behavior that humans can interpret, anticipate and coordinate with. Designing interaction policies that balance efficiency, comfort, and safety, and that work across different cultures and contexts remains difficult. While robots can be pre-programmed with specific behaviors, real-world deployments inevitably encounter situations in which these capabilities prove insufficient.{{Cite web|url= https://lcas.lincoln.ac.uk/wp/research/projects/plus-hri/ ==Societal impact and issues==