Development project
TEMPO
Temporal Context Imitation Learning for Flexible Robot Automation
TEMPO develops a new AI method enabling industrial robots to learn tasks through demonstration in hours rather than weeks of programming. By adding temporal context to imitation learning, robots can operate reliably even with varying conveyor speeds and dynamic conditions. TOS Robotics and TU Eindhoven validate this on two agrifood use cases, bell pepper packaging and meal kit assembly.
TEMPO is co-funded by the PPS Innovation Scheme for public-private partnerships of Holland High Tech | TKI HTSM.
Dutch production and packaging companies face labour shortages and rising costs. Robots could help, but current systems are too rigid. Every product change requires days to weeks of programming. Imitation learning, where an operator demonstrates the task, can break this barrier, but current methods only understand the now. They react to the current camera image without understanding recent history or anticipating what comes next. When conveyor speeds vary or objects move unpredictably, reliability drops below 50 percent. This lack of temporal awareness is the key bottleneck preventing flexible robot automation from scaling in sectors like agrifood, logistics and manufacturing.
TEMPO adds temporal context to imitation learning. The model uses not just the current image, but also recent history and dynamics prediction. This enables the robot to anticipate changing situations, similar to how an experienced worker looks ahead while packaging. TOS Robotics provides the teleoperation infrastructure, data and industrial validation. TU Eindhoven contributes state-of-the-art machine learning and scientific expertise. The result is a TRL 4 prototype on an industrial robot with conveyor belt, validated on two concrete use cases, bell pepper packaging and meal kit assembly. Target, at least 95 percent task success under varying production conditions.
TEMPO makes flexible automation feasible for SMEs currently dependent on manual labour. The technology reduces changeover times from weeks to hours and offers a payback period of less than one year. Five agrifood companies, including Bakker Barendrecht, Koninklijke Vezet and Europe Retail Packing, have signed Letters of Intent as initial customers. After the project, pilots at these companies will target commercial rollout in the agrifood packaging sector. Scientific results will be published open access and the KPI benchmark protocol will be made publicly available, ensuring broad applicability across the Dutch innovation ecosystem.
Partners
