HiSTEVE 1.0 uses a hierarchical VLA agent for open-world Minecraft
It scores 85.0% on MineStudio Bench and 37.4% average success on OpenHA, and can craft an iron pickaxe from scratch.
Stories per day · 30 days
3 RSS
Relatedrobotics4benchmarks3computer-use2minecraft2benchmark1conservation-laws1continued-pretraining1mllm1multimodal1multimodal-agents1open-world1physical-plausibility1planning1reinforcement-learning1robomme1scaling-laws1simulation1video-generation1
17 stories, newest first · one card per story
Stories tagged #embodied-agents
PAPAV framework compares computer-use and robot-use agents
Across 64 benchmarks, Act is explicitly evaluated, while only 5 assess Anticipate and 3 assess Verify.
TARSphere benchmarks long-horizon robots in changing physical and social scenes
The benchmark has 360 episodes across 30 indoor scenes, and tests of eight MLLMs found updates and human interaction still hard.
PAPAV framework compares computer-use and robot-use multimodal agents
Across 64 benchmarks, only five assess Anticipate and three assess Verify, so both remain largely hidden behind task success.
SERA self-evolving harness lifts long-horizon robot planning on RoboMME
With a fixed VLA executor, autonomous harness selection reached 51.38% success versus 44.38% without self-evolution.
Continued pre-training study scales embodied agents across virtual worlds
One hundred twenty runs span 0.8B–27B models, 50–1,600 hours of data, and five world mixtures.
PVBench scores physical plausibility in generated video
The benchmark uses 600 prompts and 46 dimensions, with object-level measurements instead of only visual-language judgments.
MC-EvoBench: Auditing Experience-Driven Self-Improvement in Minecraft Agents
MC-EvoBench tests whether Minecraft agents convert experience into durable capabilities that transfer across reset worlds.
EmbodiedCLBench: Evaluating Continual Learning for Self-Evolving Embodied Agents
EmbodiedCLBench measures whether embodied agents can recombine basic skills for harder unseen tasks.
Attacca: Goal-Directed Control under State Continuity for Long-Horizon Embodied Agents
Attacca improves long-horizon Minecraft agents to 39-47.5% clean success, up to 7x better on extended tasks.
RobotUse: Allocating Computation, Context, and Decisions
RobotUse lets robot agents revise physical actions, reaching 45% task success on RoboLab.
Reconstruct, Practice, Go Real: Guided Self-Improvement for Embodied Agents
RPG improves robot manipulation from 28.6% to 95% success in simulation and succeeds in all 30 physical trials.
Game-Guided Skill Discovery through Self-Play for Playable Agent Control
GGSD uses game self-play to learn a small set of human-playable motor skills for embodied agents.
ProgressCompass: Embodied Progress Reward Models Are Lost Without the Right Context
ProgressCompass supplies missing context so embodied progress reward models estimate long-task progress more accurately.
RoboFoundry: System-as-Policy Evolution for Self-Learning Embodied Agents
RoboFoundry evolves embodied agent systems as policies, lifting GPT-5.5 by 27.8% on EmbodiedBench.
Agent as Policy for Robotic Manipulation
Agent as Policy lets a general-purpose agent drive a physical robot via runtime reasoning and program generation, reaching 100% success on manipulation tasks.
Show-Harness: Just a VLM Agent Can Play Robots
Show-Harness enables VLM agents to control robots via a semantic action interface, achieving zero-shot frontier control and few-GPU-hour adaptation of small VLMs.