How can agentic AI be used in autonomous systems like EVs

Agentic AI refers to artificial intelligence systems with a degree of autonomy, enabling them to make decisions, take actions, and learn from experiences to achieve specific goals, often with minimal human intervention. Agentic AI systems are designed to operate independently, unlike traditional AI models that rely on predefined instructions or prompts.

Traditional AI models focus on tasks like classification or prediction. Another type of AI, generative AI, is designed to create new content, like text, images, or videos, based on patterns learned from existing data. Neither traditional AI models nor generative AI are designed for autonomous operation.

Agentic AI is an evolving technology for AI agents that enables increased autonomy based on unsupervised learning in complex environments. Agentic AI strives to mimic human cognitive and decision-making abilities.

At a high level, agentic AI consists of three elements (Figure 1):

Reinforcement learning (RL): Where the agent learns to make decisions by receiving rewards or penalties based on its actions. RL aims to identify optimal behaviors that maximize cumulative long-term rewards.
Deep neural networks (DNNs): Which use multiple layers of interconnected neurons to model complex relationships. DNNs help process unstructured data sets and make predictions without being programmed for specific scenarios.
Multi-agent systems (MAS): A key differentiator in agentic AI. They leverage several specialized agents working collaboratively and enhance the system’s ability to adapt to dynamic environments and handle complex decision-making.

Figure 1. The three keys to agentic AI performance are RL, DNNs, and MAS. (Image: E42)

Those elements support goal setting and adaptive learning. Goal-setting algorithms use a combination of knowledge about the current environment and experience to define (or refine) current goals. Adaptive learning algorithms enable agentic AI systems to apply learning from one task to related tasks, improving performance and handling new situations.

Agentic AI can leverage generative AI

Generative AI can be one of the agents in an agentic AI MAS. The agentic agents focus on autonomy and real-time decision-making in complex scenarios. The generative agents are focused on creativity and can be used to create new content like improved software code, text, or images based on learned patterns and devise new problem-solving approaches.

The generative agents could also be used to improve communication with users or other systems (Figure 2).

Figure 2. Agentic AI and generative AI are complimentary approaches, and a generative agent can be included in an agentic AI MAS. (Image: FabriXAI)

The emergence of agentic AI is following a path of increasing performance. It’s growing from today’s base of AI-augmented automation, Level 1 in Figure 3, in a series of four additional steps, beginning with the addition of the ability to determine the intent and outcome of processes. Next is planning and adapting to changes, followed by the ability to self-refine and update instructions without outside intervention. Full autonomy requires creativity and the ability to anticipate changing needs before they occur proactively.

Figure 3. This multi-step road map has been outlined for the development of agentic AI. (Image: K21 Academy)

The key to autonomy

The deployment of agentic AI outlined in Figure 3 is analogous to the Automotive Advanced Driver-Assistance Systems (ADAS) categories outlined by the Society of Automotive Engineers. ADAS categories range from Level 0 without automation to Level 5 for fully autonomous driving. Agentic AI will be key to arriving at ADAS Level 5 performance. Three of the anticipated benefits of agentic AI in automobiles include:

Optimal navigation and path planning for improved energy efficiency.
Dynamic and proactive adaptation to changing conditions in real-time.
Hazard detection and avoidance for improved safety.

Agentic AI will also benefit Industry 4.0 facilities. It monitors machinery in real time, predicts failures, schedules maintenance, reduces downtime, and optimizes asset availability. It will enable continuous process optimization, minimize waste, and enhance operational efficiency.

Summary

Agentic AI is an evolving technology expected to lead to full autonomy in applications like automobiles and industry 4.0 factories. MAS is a key differentiator of agentic AI and leverages multiple agents’ strengths to enhance performance. The path to full autonomy for agentic agents is analogous to the path for Level 5 ADAS performance outlined by the SAE.