Cognitive robotic abstract machine
Cognitive robot abstract machine (CRAM) is a cognitive architecture for autonomous robots designed to perform everyday manipulation tasks in human environments.[1][2] It was developed by Beetz et al. at the University of Bremen. It claims to let robots transform underspecified task instructions into context‑sensitive and executable actions.[3] This is based on its current beliefs about the environment and the predicted outcomes of actions.[3]
Description
[edit]CRAM is a hybrid cognitive architecture that integrates symbolic and sub‑symbolic representations and processes.[3] It can achieve this by transforming underspecified abstract task instructions into context‑sensitive executable actions.[1]
The architecture connects symbolic knowledge directly to both the robot's perception and movement systems and to the internal data structures that govern its behavior.[4] The core of the system is built around two primary elements, which are the CPL plan language and the KnowRob knowledge processing framework.[4][5]
History
[edit]Beetz et al. introduced CRAM at the 2010 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), where they described it as a software toolbox for cognition‑enabled robotics.[1][5] It has been extended with features like semantic digital twins, episodic memory systems, and dual‑process reasoning.[3] The second generation, CRAM 2.0, was developed as part of a German Research Foundation (DFG) project and was introduced in a 2025 article in the journal Cognitive Systems Research.[6][7]
Architecture
[edit]CRAM is a cognitive architecture that employs generalized action plans.[3] They are transformed into parameterized low‑level motion plans using knowledge and reasoning with a contextual model.[6] Process modules can be configured by the higher‑level control logic. These modules interpret the properties of designators and convert them into specific low-level commands based on the robot's current understanding of its environment.[4][1]
KnowRob is a knowledge processing system that delivers reasoning and knowledge services for autonomous robots.[4] Its modular design can allow components to be loaded only when required.[4]
Applications
[edit]CRAM has been used in research on cognition‑enabled robot control, including mobile manipulation in household environments.[1][8] The CRAM cognitive architecture allows a robot to carry out activities such as laying a table for a meal and loading a dishwasher afterward.[2]
Differences between other architectures
[edit]CRAM is one of several cognitive architectures for robotics.[3][9] According to Beetz and Kümpel et al., its core feature was its emphasis on lightweight reasoning.[3] A 2025 study by Beetz et al. on CRAM 2.0 shows that the architecture addresses underdetermined task specification through generalized action plans that are instantiated via contextual queries to a semantic knowledge base.[6]
See also
[edit]References
[edit]- 1 2 3 4 5 Beetz, Michael; Mösenlechner, Lorenz; Tenorth, Moritz (2010). "CRAM – A Cognitive Robot Abstract Machine for Everyday Manipulation in Human Environments". IEEE/RSJ International Conference on Intelligent Robots and Systems. pp. 1012–1017. doi:10.1109/IROS.2010.5650146.
- 1 2 Vernon, David (2022). "Action Selection and Execution in Everyday Activities: A Cognitive Robotics and Situation Model Perspective". Topics in Cognitive Science. 14 (2): 344–362. doi:10.1111/tops.12569. PMID 34459566.
- 1 2 3 4 5 6 7 Beetz, Michael; Kümpel, Michaela (2026). "From Frames to Pouring: The CRAM Cognitive Architecture for Everyday Robot Manipulation". KI – Künstliche Intelligenz. doi:10.1007/s13218-026-00917-z.
- 1 2 3 4 5 "CRAM Cognitive Robot Abstract Machine". University of Bremen. Retrieved 2026-08-29.
- 1 2 Vernon, David (2022). "Cognitive Architectures" (PDF). Cognitive Robotics. MIT Press. pp. 191–212.
- 1 2 3 Beetz, Michael; Kazhoyan, Gayane; Vernon, David (2025). "Robot manipulation in everyday activities with the CRAM 2.0 cognitive architecture and generalized action plans". Cognitive Systems Research. 92 101375. doi:10.1016/j.cogsys.2025.101375.
- ↑ "CRAM 2.0 — 2. Generation einer kognitiven Architektur für die Ausführung von alltäglichen Manipulationsaufgaben". GEPRIS. German Research Foundation. Retrieved 2026-08-29.
- ↑ Evaluation of Cognitive Architectures: SASE vs. CRAM (Report). Karlsruhe Institute of Technology. 2024.
- ↑ Reip, Michael (2016). Dependable Belief Management for High-Level Robot Programs (PDF) (Master's thesis). Graz University of Technology.