Meta-World
A lightweight meta-reinforcement-learning benchmark containing 50 tabletop robotic arm manipulation tasks. It is commonly used to test the generalization and multi-task learning capabilities of policies.
Timeline (last 90 days)
HexaShell achieved a 62% success rate on the Meta-World door-opening task.
Current benchmarks (RLBench and Meta-World) lean toward single-arm fixed-base setups, whereas industrial production lines often use 7-DOF arms or mobile bases, so conclusions should be transferred with caution.
Meta-World is a lightweight meta-reinforcement-learning benchmark containing 50 tabletop robot-arm manipulation tasks, commonly used to test policy generalization and multi-task learning ability.
Meta-World provides 50 tabletop manipulation tasks of varying difficulty, RLBench includes 100 tasks and supports multiple observation modes, and CALVIN focuses on language-conditioned long-horizon manipulation task chains — these are the main simulation evaluation benchmarks in the field of robotic manipulation.
Benchmark platforms such as Meta-World, RLBench, and CALVIN provide standardized multi-task evaluation environments.
All Facts (5)
HexaShell achieved a 62% success rate on the Meta-World door-opening task.
50%UnverifiedCurrent benchmarks (RLBench and Meta-World) lean toward single-arm fixed-base setups, whereas industrial production lines often use 7-DOF arms or mobile bases, so conclusions should be transferred with caution.
50%UnverifiedMeta-World is a lightweight meta-reinforcement-learning benchmark containing 50 tabletop robot-arm manipulation tasks, commonly used to test policy generalization and multi-task learning ability.
50%UnverifiedMeta-World provides 50 tabletop manipulation tasks of varying difficulty, RLBench includes 100 tasks and supports multiple observation modes, and CALVIN focuses on language-conditioned long-horizon manipulation task chains — these are the main simulation evaluation benchmarks in the field of robotic manipulation.
50%UnverifiedBenchmark platforms such as Meta-World, RLBench, and CALVIN provide standardized multi-task evaluation environments.
50%