AI
Risk-Averse Online POMDP Planning via CVaR of the Immediate Cost with Performance Guarantees
Publication: cs.AI updates on arXiv.org
Author: Yaacov Pariente, Vadim Indelman
Source published: 30/09/2026, 04:00:00 UTC
Briefing created: 30/09/2026, 04:18:11 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 30/09/2026, 04:03:24 UTC.
Briefings
This paper proposes a method for online POMDP planning that uses Conditional Value at Risk (CVaR) applied to the immediate cost over the belief, rather than the value function. The authors retain the standard expected cumulative return as the objective, transforming the problem into a standard MDP structure where only the cost computation is modified. A key theoretical result is a finite-time bound on the gap between a particle belief MDP surrogate and the original POMDP, providing end-to-end guarantees for policy evaluation and sparse sampling. The method achieves risk-sensitive planning with finite-time guarantees for policy evaluation and sparse sampling, with estimation error independent of the risk level. In the risk-neutral limit, the formulation recovers standard expectation-based planning.
Key points
- The paper applies CVaR to the immediate cost over the belief at each step to address risk within the belief, unlike prior methods that modify the value function.
- The standard expected cumulative return is retained as the objective, resulting in a standard MDP structure with only the cost computation modified for risk sensitivity.
Context and limitations
The study focuses on online POMDP planning with CVaR applied to immediate cost; it does not address other risk-sensitive objectives or alternative value function modifications.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 30/09/2026, 04:18:11 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeiftcowl5cnvokk2hqan3i77gcrf56whoqnxulwn62hcxwg7gpiamy