登录查看更多内容

What is the best way to ensure a robotics algorithm can handle unexpected situations?

由人工智能和领英社区提供技术支持

Robots are becoming more capable and autonomous, but they still face many challenges in the real world. Unexpected situations, such as changes in the environment, failures of sensors or actuators, or human interference, can disrupt their normal operation and cause errors or accidents. How can you ensure that your robotics algorithm can handle these situations gracefully and safely? In this article, you will learn some of the best practices and techniques to design, test, and improve your robotics algorithm for robustness and adaptability.

此文章中的业界达人

由社区从 8 条内容中精选。了解更多

Dharunkumar Senthilkumar

| Machine Learning, Robotics and Control | MSc MPSYS at Chalmers University |
Varun Nayak

Robotics @ Dexterity, Inc. | Stanford | BITS
Marco Roggero

Senior Application Engineer at MathWorks

1 Define your goals and assumptions

The first step to ensure your robotics algorithm can handle unexpected situations is to define your goals and assumptions clearly. What is the purpose of your algorithm? What are the expected inputs and outputs? What are the constraints and requirements? What are the assumptions you make about the environment, the robot, and the human? By answering these questions, you can specify the scope and the performance criteria of your algorithm, and identify the potential sources of uncertainty and variability.

添加您的观点

Ernesto Martínez de Carvajal Hedrich

Prof. Univ. en FNB, Tecnocampus e Inesdi Digital Business School - Autor libros Robótica Educativa, Inteligencia Artificial, Náutica y Aeronáutica - Perito Judicial Informático - Comisario de Averías
举报内容
By definition, an "unforeseen situation" is one that has not been taken into account when designing the algorithm and, therefore, it is impossible for it to manage it adequately, beyond using a generic routine for these cases, such as stopping the process or requesting help from the human ("I'll move on to the operator"). Talking, for example, about machine learning systems implies that these possible situations have been taken into account and, if the algorithm is well designed, it will not "unexpected situations" can occur and, if they occur, we return to the initial point: they have not been taken into account when designing the algorithm.

已翻译

赞
Peter Struwing

Electro-Mechcanical Engineering Technologist | Life Long Industrial Automation Specialist | Project Professional | Automotive & Aircraft Enthusiast
举报内容
All unexpected situations must immediately stop all robots, this is an Emergency Stop situation. The robot cannot move without human intervention and must follow the humans commands until the situation is confirmed resolved by the human.

已翻译

赞
Kausik P(He,Him,His)

Robotics | ROS | Product Design | Mechanical Engineer
举报内容
Understand what are the edge cases you need to cover. There might be some unique cases in which you might take a few special considerations. Make separate test cases for it so that it would be taken care during development.

已翻译

赞

2 Use probabilistic models and methods

The second step to ensure your robotics algorithm can handle unexpected situations is to use probabilistic models and methods. Probabilistic models and methods allow you to represent and reason about the uncertainty and variability in the data, the environment, and the robot. For example, you can use Bayesian networks, Markov models, or Kalman filters to model the state of the robot and the environment, and update them based on the sensor measurements and the actions. You can also use probabilistic methods to plan and execute actions that maximize the expected utility or minimize the expected risk.

添加您的观点

3 Incorporate feedback and learning

The third step to ensure your robotics algorithm can handle unexpected situations is to incorporate feedback and learning. Feedback and learning enable your algorithm to adapt and improve over time, based on the outcomes of the actions and the interactions with the environment and the human. For example, you can use reinforcement learning, imitation learning, or active learning to learn from rewards, demonstrations, or queries, and adjust your algorithm accordingly. You can also use feedback and learning to detect and correct errors, recover from failures, or explore new situations.

添加您的观点

Varun Nayak

Robotics @ Dexterity, Inc. | Stanford | BITS
举报内容
Log everything! Whether you like it or not, a robotic system is a giant hairy ball of unobservable global state. The path towards building a very robust robotic system must include a good (often structured) logging methodology that unlocks rapid learning/iteration in the development process through high visibility. Don't skimp on log statements, especially during the initial phases of the product life cycle.

已翻译

赞
Kausik P(He,Him,His)

Robotics | ROS | Product Design | Mechanical Engineer
举报内容
Make a feedback loop or try to compile the data during operation you can use RL to make the robot learn it on the go. If your robot uses ROS, you can do it easily using the rosbag command.

已翻译

赞

4 Test and evaluate your algorithm

The fourth step to ensure your robotics algorithm can handle unexpected situations is to test and evaluate your algorithm. Testing and evaluation allow you to verify and validate your algorithm, and measure its performance and robustness. For example, you can use simulation, experimentation, or user studies to test your algorithm in different scenarios, with different levels of uncertainty and variability, and with different types of users. You can also use metrics, such as accuracy, reliability, efficiency, or usability, to evaluate your algorithm against your goals and assumptions.

添加您的观点

Dharunkumar Senthilkumar

| Machine Learning, Robotics and Control | MSc MPSYS at Chalmers University |
举报内容
here are the important points you can look into for evaluating your algorithm 1. Benchmark your algorithm against existing baselines or state-of-the-art methods. This provides context for understanding the strengths and weaknesses of your approach and facilitates comparison with industry standards. 2. Test your algorithm against edge cases and corner cases. These are scenarios that lie at the extremes or boundaries of your system's capabilities. Identifying how well your algorithm handles such cases is crucial for robustness. 3. Evaluate the long-term stability of your algorithm. Assess whether its performance degrades over time due to factors like sensor drift, environmental changes, or wear and tear of hardware components.

已翻译

赞

5 Follow ethical and safety standards

The fifth step to ensure your robotics algorithm can handle unexpected situations is to follow ethical and safety standards. Ethical and safety standards provide guidelines and principles for designing, developing, and deploying your algorithm, and ensuring its compliance with the laws, regulations, and values of the society. For example, you can follow the IEEE Code of Ethics for Robotics Engineers, or the Asilomar AI Principles, to ensure that your algorithm respects the human dignity, rights, and interests, and avoids harm, bias, or deception.

添加您的观点

Marco Roggero

Senior Application Engineer at MathWorks
举报内容
SOTIF (Safety Of The Intended Functionality) pertains to ensuring a system operates safely as intended, even in the absence of malfunctions. It's essential in the automotive sector, particularly for autonomous vehicles, and is outlined in ISO/PAS 21448. SOTIF is equally valuable in robotics. Robots operate in unpredictable environments, and it's crucial they perform safely. Applying SOTIF principles helps identify and mitigate risks not caused by failures but by limitations in design or unexpected conditions. Key benefits for robotics include Risk Mitigation,Reliability, Trust and Compliance.

已翻译

赞

6 Keep learning and improving

The sixth and final step to ensure your robotics algorithm can handle unexpected situations is to keep learning and improving. Learning and improving enable you to update and enhance your algorithm, and cope with the changing and evolving needs and expectations of the users and the society. For example, you can use online learning, transfer learning, or lifelong learning to adapt your algorithm to new data, tasks, or domains. You can also use feedback loops, user feedback, or peer review to identify and address the limitations, challenges, or opportunities of your algorithm.

添加您的观点

7 Here’s what else to consider

This is a space to share examples, stories, or insights that don’t fit into any of the previous sections. What else would you like to add?

添加您的观点

Alaa Serhal

Senior Product Manager | Test Automation at Murex
举报内容
Ensuring a robotics algorithm can handle unexpected situations involves a combination of robust design principles, comprehensive testing, and adaptive strategies. One of the concepts to add to this article is 'Simulation', you can test the implemented algorithm and see the outcome in stressful and unexpected situations. You can define endless parameters that can resonate what can and can't happen, while minimizing costs and waste.

已翻译

赞

Robotics

+ 关注

给文章评分

我们借助人工智能创建了此文章。您认为这篇文章怎么样？

很棒不太好

举报此文章

查看全部

What is the best way to ensure a robotics algorithm can handle unexpected situations?

1

2

3

4

5

6

7

1 Define your goals and assumptions

2 Use probabilistic models and methods

3 Incorporate feedback and learning

4 Test and evaluate your algorithm

5 Follow ethical and safety standards

6 Keep learning and improving

7 Here’s what else to consider

Robotics

给文章评分

感谢您的反馈

更多Robotics相关文章

更多相关阅读内容

What is the best way to ensure a robotics algorithm can handle unexpected situations?

1

2

3

4

5

6

7

1 Define your goals and assumptions

2 Use probabilistic models and methods

3 Incorporate feedback and learning

4 Test and evaluate your algorithm

5 Follow ethical and safety standards

6 Keep learning and improving

7 Here’s what else to consider

Robotics

给文章评分

感谢您的反馈

查看其他技能