Chart-RL: Generalized Chart Comprehension via Reinforcement Learning with Verifiable Rewards

Xin Zhang; Xingyu Li; Rongguang Wang; Ruizhong Miao; Zheng Wang; Dan Roth; Chenyang Li

cs.LGMar 7, 2026

Chart-RL: Generalized Chart Comprehension via Reinforcement Learning with Verifiable Rewards

Authors: Xin Zhang, Xingyu Li, Rongguang Wang, Ruizhong Miao, Zheng Wang, Dan Roth, Chenyang Li

Organizations: Oracle AI

Abstract

Accurate chart comprehension represents a critical challenge in advancing multimodal learning systems, as extensive information is compressed into structured visual representations. However, existing vision-language models (VLMs) frequently struggle to generalize on unseen charts because it requires abstract, symbolic, and quantitative reasoning over structured visual representations. In this work, we introduce Chart-RL, an effective reinforcement learning (RL) method that employs mathematically verifiable rewards to enhance chart question answering in VLMs. Our experiments demonstrate that Chart-RL consistently outperforms supervised fine-tuning (SFT) across different chart understanding benchmarks, achieving relative improvements of 16.7% on MutlChartQA, and 11.5% on ChartInsights. We conduct robustness analysis, where Chart-RL achieves enhanced performance in 18 of 25 perturbed chart categories, demonstrating strong consistency and reasoning capability across visual variations. Furthermore, we demonstrate that task difficulty and inherent complexity are more critical than data quantity in RL training. For instance, Chart-RL trained on merely 10 complex chart-query examples significantly outperforms models trained on over 6,000 simple examples. Additionally, training on challenging reasoning tasks not only improves in-domain generalization relative to simpler tasks, but also facilitate strong transfer to out-of-domain visual mathematical problems.

Figures & tables

Figure 1: Comparison of Chart-RL vs SFT to reason on complex chart question answering.

Figure 2: Schematic diagram of Chart-RL framework with GRPO optimization for chart comprehension.

Method	Training Task	MultiChartQA Overall			ChartInsights Reasoning
		Accuracy	Rel. $\Delta$	$p$ -value	Accuracy	Rel. $\Delta$	$p$ -value
Baseline	–	35.2	–	–	36.4	–	–
SFT	Hard	32.9	-6.6%	0.12	34.7	-4.7%	$<10^{-2}$
CoT-SFT		39.1	+11.1%	$<10^{-2}$	40.3	+10.7%	$<10^{-6}$
Chart-RL(ours)		41.1	+16.7%	$<10^{-4}$	40.6	+11.5%	$<10^{-6}$

Table 1: Evaluation results on two benchmarks comparing our Chart-RL approach against baseline, SFT, and CoT-SFT. The rightmost column for each dataset shows the proportional test results for the relative change.

Category	Dataset	# Charts	# Queries
Training	CharXiv	448	448
	PlotQA	6,200	6,200
Evaluation	RobustCQA	400	400
	MultiChartQA	1,370	2,000
	ChartInsights	2,000	22,000

Table 2: Statistics of training and evaluation datasets.

Category	Fine-Tuning Strategy
Category	SFT	Chart - RL (ours)
annotations	0.41	0.41
benchmark	0.47	0.52
color_random	0.37	0.37
color_scheme	0.34	0.34
data_pivot	0.35	0.36

Table 3: Robustness comparison between Chart-RL and SFT across 25 perturbed chart categories.

Method	Training Task	Training Samples	MultiChartQA Overall			ChartInsights Reasoning
			Accuracy	Rel. $\Delta$	$p$ -value	Accuracy	Rel. $\Delta$	$p$ -value
Baseline	–	–	35.2	–	–	36.4	–	–
Chart-RL (ours)	Hard	10	39.1	+11.1%	$<10^{-2}$	39.1	+7.4%	$<10^{-4}$
		100	39.2	+11.4%	$<10^{-2}$	39.7	+9.1%	$<10^{-6}$
		448	41.1	+16.7%	$<10^{-4}$	40.6	+11.5%	$<10^{-6}$

Table 4: Evaluation results for Chart-RL across different training sample sizes, showing performance on three chart understanding benchmarks under varying RL training data scales. The rightmost column for each dataset shows the proportional test results for the relative change.

Figure 3: RL Training reward trajectories for Chart-RL across different training sample sizes, showing accuracy rewards (left) and format rewards (right) throughout the training process.

Methods	MathVerse Vision
	Acc.	Rel. $\Delta$	$p$ -value
Baseline	1.8	–	–
Chart-RL (ours)	2.8	+55.6%	$<10^{-2}$

Table 5: Evaluation results on MathVerse Vision.

Figure 4: Performance breakdown by sub-category on MultiChartQA and ChartInsights benchmarks.

Figure 5: Accuracy reward trajectories in RL training for Easy and Hard tasks.

Task	MultiChartQA	ChartInsights
Baseline	35.2	36.4
Easy	34.8 (-1.2%)	36.0 (-1.1%)
Hard	41.1 (+16.8%)	40.6 (+11.5%)

Table 6: Evaluation results of Easy vs. Hard training tasks on different benchmarks.

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 6: An example of easy task for RL training.

Figure 7: An example of hard task for RL training.

Chart QA

Chart question answering (Chart QA) develops systems that read charts and answer questions requiring visual, numerical, and comparative reasoning.

See all 32 papers

Chart-RL: Policy Optimization Reinforcement Learning for Enhanced Visual Reasoning in Chart Question Answering with Vision Language Models
Apr 3, 2026Yunfei Bai, Amit Dhanda, Shekhar JainChart QA VLM Reasoning
Amazon
CharTool: Tool-Integrated Visual Reasoning for Chart Understanding
Apr 3, 2026Situo Zhang, Yifan Zhang, Zichen Zhu +6Chart Understanding Chart QA
X-LANCE Lab, Shanghai Jiao Tong University, Shanghai, China · East China Normal University, Shanghai, China · Shanghai Innovation Institution, Shanghai, China +1
ChartProbe: A Diagnostic Study on Visual Reasoning through Perception, Grounding, and Simple Reasoning
Aug 13, 2026Mahsa Khoshnoodi, Sarah Adel BargalChart Understanding Chart QA
Department of Computer Science, Georgetown University

Apr 3, 2026J/K move · Enter open · S save

Chart-RL: Policy Optimization Reinforcement Learning for Enhanced Visual Reasoning in Chart Question Answering with Vision Language Models

Yunfei Bai, Amit Dhanda, Shekhar Jain

Amazon

Chart QA VLM Reasoning RL for VLMs

Open paper AskPDF arXiv

The recent advancements in Vision Language Models (VLMs) have demonstrated progress toward true intelligence requiring robust reasoning capabilities. Beyond pattern recognition, linguistic reasoning must integrate with visual comprehension, particularly for Chart Question Answering (CQA) tasks involving complex data visualizations. Current VLMs face significant limitations in CQA, including imprecise numerical extraction, difficulty interpreting implicit visual relationships, and inadequate attention mechanisms for capturing spatial relationships in charts. In this work, we address these challenges by presenting Chart-RL, a novel reinforcement learning framework that enhances VLMs chart understanding through feedback-driven policy optimization of visual perception and logical inference. Our key innovation includes a comprehensive framework integrating Reinforcement Learning (RL) from Policy Optimization techniques along with adaptive reward functions, that demonstrates superior performance compared to baseline foundation models and competitive results against larger state-of-the-art architectures. We also integrated Parameter-Efficient Fine-Tuning through Low-Rank Adaptation (LoRA) in the RL framework that only requires single GPU configurations while preserving performance integrity. We conducted extensive benchmarking across open-source, proprietary, and state-of-the-art closed-source models utilizing the ChartQAPro dataset. The RL fine-tuned Qwen3-VL-4B-Instruct model achieved an answer accuracy of 0.634, surpassing the 0.580 accuracy of the Qwen3-VL-8B-Instruct foundation model despite utilizing half the parameter count, while simultaneously reducing inference latency from 31 seconds to 9 seconds.

Abstract

Chart QA

Explore similar work

Chart-RL: Policy Optimization Reinforcement Learning for Enhanced Visual Reasoning in Chart Question Answering with Vision Language Models

CharTool: Tool-Integrated Visual Reasoning for Chart Understanding

ChartProbe: A Diagnostic Study on Visual Reasoning through Perception, Grounding, and Simple Reasoning