Authored by  and Felix-Olivier Duguay

Introduction

Qualities of perception and planning are one way to recognize intelligence in animals. Unlike manufacturing robots that are designed to follow a pre-specified sequence of actions, the smart robots can predict the effect of a sequence of actions using knowledge of the environment and adapt according to their objectives. This intelligence could be achieved by merging the well-known field of Artificial Intelligence (AI) and robotics. In this article, a review of AI and its advances are provided, followed by a demonstration of how robotics and AI communities can benefit from cooperating with each other to overcome current challenges. A discussion will follow about how this collaboration could be more efficient and cost-effective if the computation and storage tasks are offloaded to Cloud Robotics as the engine to perform complex tasks virtually instead of investing in equipping ordinary robots. 

What is Artificial Intelligence

Artificial intelligence is the field of computer science that focuses on making machines smart and able to come up with inferences, assumptions, and conclusions in non-trivial situations. AI is necessary to enable machines to fulfill the chores that humans cannot complete due to danger, threat, inaccessibility, or complexity of the problem. Nowadays, AI is used in many applications, which includes but is not limited to gaming, speech recognition, computer vision, expert systems, and robotics. [1]  

Computer vision, a branch of AI that has shown great progress in the last decade, is very useful when controlling a mobile robot or to enhance the capability of optical sensors. Object recognition and simultaneous localization and mapping (SLAM) are two key tools in this regard. Either for an unmanned aerial vehicle (UAV), a service robot or a surveillance camera, it is vital to detect obstacles and moving objects (object recognition). Navigating machines would also require knowing their exact location by generating their traveling map (SLAM). 

Artificial Intelligence and Robotics Collaboration

The robotics industry has traditionally concentrated its focus on sensors, controls, and implementation of the appropriate actions such as the motion of an arm or a wheel precisely, without collision. Yet, the AI community has conventionally focused on more complex problems with massive computations, requiring a more intelligent grasp of the environment and comprehensive analysis of the input sensor data to determine an upcoming sequence of actions. [2] 

In order to transition robots from their current industrial use to intelligent mobile agents, robotics and AI communities are cooperating more to benefit from their respective breakthroughs. The robotics industry will benefit from such collaboration to merge advanced sensing and computations with their existing not-necessarily-intelligent robots. In other words, by integrating AI modules in robots, more complex and organized tasks can be accomplished by robots, especially in human unfriendly zones such as hazardous environments, military missions, remote areas, etc. Due to the complex perception and planning AI modules, robots can notice changes in the environment more efficiently and react with more agility and precision. The AI community, on the other hand, has the chance to expand its field of research into diverse robotic applications and industries. This benefits the AI ecosystem: test and assess its modules in many different fields by controlling intelligent robots. 

Robotics challenges to be resolved by Artificial Intelligence

Integrating AI into robotics is an encouraging approach and an active interest for both fields. The robotics industry is facing multiple challenges that AI can facilitate overcoming. The challenges are well investigated in [3] and a brief review follows.  

  • Learning, analyzing, and responding to a new and complex environment. This is useful when the state is dynamic and not fully observable. In such scenarios, AI can be integrated to enable a more agile response. 
  • Applying control policies in multi-robot systems. Examples of such actions could be controlling swarm robots or multilayer control systems in scenarios where massive amounts of data are available, but no state model is known.  
  • Advanced manipulation. This is regarding the applications that require grasping or using the tools in novel situations. 
  • Advanced Object Recognition. This goes a step further than recognizing the objects, but perceiving the classification of the objects to be able to use them efficiently, even in unknown schemes and applications. 
  • Perceiving and anticipating human behavior. Robots can foresee human behavior in scenarios where they interact with the public.  
  • Sensor fusion. This applies to massive sensor networks where the abundant data received from sensors is gathered and analyzed to create meaningful and reliable inferences using combined knowledge from the different sources. 
  • Advanced task planning. This challenge deals with planning and executing the above-mentioned tasks in an efficient and reliable ordered sequence to achieve the required objective. 

Cloud Robotics

Adopting AI in robotics is limited by the processing and storage capacities of the robots. The performance could suffer from the lack of storage or processing power. Offloading the processes and storage to the clouds is a promising trend in the industry. Cloud robotics enables transferring the computation-intensive tasks to server infrastructures, providing more agile processing on performant parallel processors in the clouds. The centralized processing provided by cloud robotics makes it easier to learn from experience by treating the abundant data received from multi-robots all at once. Moreover, communication and data transfer for multi-robots in swarm robotics is very natural since the data is centralized. Robots can complete their tasks while cooperating through clouds to enhance their sensors. [4] 

In a nutshell

Merging the advanced features of AI into robotics would expand the robots’ functionalities from simple state machines to smart gadgets capable of recognition, analysis, and response. The robots could interact in automated tasks without the need to be controlled or supervised. Yet, due to the intensive computation and data storage required for AI applications, the robots constantly need to be upgraded to avoid latency and performance degradation. Cloud robotics is a promising trend to provide the agility and efficiency of AI in robotics industries. The not-necessarily-intelligent robots would be able to handle tasks much more difficult than permitted by their built-in hardware by offloading their data storage and processing to the cloud infrastructure. In addition, the AI applications are able to process data from different sources all at once to enhance the robots’ perception and improve the quality of decisions. 

In case you missed it, C2RO Cloud Robotics was featured as 1 of 4 tech companies leading the AI surge in Montreal.

References: 

  1. What is AI. Link 
  2. L.P. Kaelbling, “Artificial intelligence and robotics,” in Thirty-Third IEEE Computer Society International Conference, Digest of Papers Compcon, 1988, pp.59-61 
  3. H.A, Pierson, M.S. Gashler, “Deep Learning in Robotics: A Review of Recent Research,” in Journal of Computing Research Repository (CoRR), 2017. 
  4. Why Cloud Robotics? Link 

Build Smarter Robots

Join the beta by entering your email address

Thank you for signing up for the C2RO Beta program. Within 24-48 hours you'll be contacted with next steps for joining.

Build Smarter Robots with C2RO

Join the beta with your email address

Thank you for signing up for the C2RO Beta. Within 24-48 hours you'll be contacted with next steps for joining.

Sign Up for Beta Access

The C2RO Cloud Robotics platform is ready today. Sign up and one of our developers will be in touch shortly to help you get started.

You have successfully signed up!