All Tools
C
AgentsFreeOpen Source
CALVIN
Language-conditioned long-horizon robot manipulation benchmark
MIT
ABOUT
Language-conditioned robot agents need long-horizon tasks, not single-step pick-and-place demos. CALVIN gives a shared simulated kitchen with multi-step language goals, onboard cameras, and tactile options so researchers train and compare policies that follow human instructions over many sequential actions.
INTEGRATION GUIDE
1. Train language-conditioned manipulation policies on CALVIN's D, ABC, or ABCD splits
2. Evaluate long-horizon instruction following with onboard RGB and tactile sensors
3. Benchmark multi-task robot agents that chain open, pick, and place skills
4. Prototype vision-language robot policies before transferring to real hardware
TAGS
pythonagentsroboticslanguage-conditionedmanipulationsimulationbenchmark