---
title: "PAC-MAN: Perception-Aware CBF-RL for Whole-Body Safety in Humanoid Dodgeball"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.28623v1"
permalink: "https://feed.astralor.com/items/itm_ca679b7835c324fca5e61440"
published: "2026-07-30T17:59:35.000Z"
score: 76
channel: "ai"
tags: ["PAC-MAN","Unitree","Humanoid Robots","Reinforcement Learning","Perception","Control Barrier Function"]
---

# PAC-MAN: Perception-Aware CBF-RL for Whole-Body Safety in Humanoid Dodgeball

## AI commentary

> What excites me most is that this paper liberates the safety layer from the assumption of omniscient perception and confronts real sensor noise. For engineers deploying robots, a 95% real-world dodgeball success rate is a hard metric, proving the method isn't just simulation-pretty. If you work on safety-critical RL or humanoid behavior control, this is worth a deep read.

## At a glance

This paper presents PAC-MAN, a perception-aware CBF-RL framework that couples control-barrier safety with deployment-realistic onboard sensing for whole-body humanoid dodgeball. The policy uses only segmentation-masked depth from a head-mounted camera, with training-time CBF guidance for clearance to every body link and an adversarial motion prior to regularize evasive reflexes. Evaluated on a controlled any-link contact benchmark with seeded throws, the policy approaches a privileged state oracle in both single-throw and deployment-loop regimes. The authors find that barrier structure depends on perceptual observability: Joint-CBF performs best with accurate ball states, degrades under fixed-camera observations when used only as training guidance, and recovers with a ball-tracking gimbal or privileged runtime filter. They deploy a lightweight Link-CBF policy zero-shot on the Unitree G1 in the real world, achieving 95% throw evasion success and semantic segmentation to dodge different balls.

## Key points

- Proposes PAC-MAN, coupling CBF safety constraints with realistic perception for whole-body humanoid dodgeball.
- Uses only segmentation-masked depth from a head-mounted camera, approaching privileged oracle performance.
- Barrier structure depends on perceptual observability; Joint-CBF degrades under fixed-camera observations but recovers with a ball-tracking gimbal.
- Zero-shot deployment on Unitree G1 achieves 95% real-world throw evasion, with semantic segmentation for different balls.

## Full text

We present PAC-MAN, a perception-aware CBF-RL framework that couples control-barrier safety with deployment-realistic onboard sensing for whole-body humanoid dodgeball. The deployed policy sees the ball only as segmentation-masked depth from a head-mounted camera, while training-time CBF guidance represents clearance to every body link, and an adversarial motion prior regularizes the resulting evasive reflexes. We evaluate on a controlled any-link contact benchmark with seeded throws in two regimes: single throws and a deployment loop in which the robot walks back to its station and recovers between throws. On this benchmark, the policy comes within a few points of a privileged state oracle: a fixed onboard camera alone is adequate for evasion. We find that usable barrier structure depends on perceptual observability: Joint-CBF gives the best performance with accurate ball states, degrades under fixed-camera observations when used only as training guidance, and recovers with a ball-tracking gimbal or privileged runtime filter. We therefore deploy a lightweight Link-CBF policy zero-shot on the Unitree G1 in the real world, where it tolerates imperfect perception, succeeds on 95% of throws, and uses semantic segmentation to dodge different balls.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_ca679b7835c324fca5e61440