VLA

Search documents
TransDiffuser: 理想VLA diffusion出轨迹的架构
理想TOP2· 2025-05-18 13:08
Core Viewpoint - The article discusses the advancements in the field of autonomous driving, particularly focusing on the Diffusion model and its application in generating driving trajectories, highlighting the differences between VLM and VLA systems [1][4]. Group 1: Diffusion Model Explanation - Diffusion is a generative model that learns data distribution through a process of adding noise (Forward Process) and removing noise (Reverse Process), akin to a reverse puzzle [4]. - The model's denoising process involves training a neural network to predict and remove noise, ultimately generating target data [4]. - Diffusion not only generates the vehicle's trajectory but also predicts the trajectories of other vehicles and pedestrians, enhancing decision-making in complex traffic environments [5]. Group 2: VLM and VLA Systems - VLM consists of two systems: System 1 mimics learning to output trajectories without semantic understanding, while System 2 has semantic understanding but only provides suggestions [2]. - VLA is a single system with both fast and slow thinking capabilities, inherently possessing semantic reasoning [2]. - The output of VLA is action tokens that encode the vehicle's driving behavior and surrounding environment, which are then decoded into driving trajectories using the Diffusion model [4][5]. Group 3: TransDiffuser Architecture - TransDiffuser is an end-to-end trajectory generation model that integrates multi-modal perception information to produce high-quality, diverse trajectories [6][7]. - The architecture includes a Scene Encoder for processing multi-modal data and a Denoising Decoder that utilizes the DDPM framework for trajectory generation [7][9]. - The model employs a multi-head cross-attention mechanism to fuse scene and motion features during the denoising process [9]. Group 4: Performance and Innovations - The model achieves a Predictive Driver Model Score (PDMS) of 94.85, outperforming existing methods [11]. - Key innovations include anchor-free trajectory generation and a multi-modal representation decorrelation optimization mechanism to enhance trajectory diversity and reduce redundancy [11][12]. Group 5: Limitations and Future Directions - The authors note challenges in fine-tuning the model, particularly the perception encoder [13]. - Future directions involve integrating reinforcement learning and referencing models like OpenVLA for further advancements [13].
理想汽车,压力山大
虎嗅APP· 2025-05-09 13:14
出品丨虎嗅汽车组 作者丨肖漫 头图丨视觉中国 尽管账上有千亿现金,但理想还没有迎来躺着数钱的时候。 理想今年的销量目标是70万辆,比去年多了20万,而理想规划的新车型仅有理想i8和理想i6两款,且都在下半年发布,其他均为改款焕新。销量的重 担,依然落在L系列上。然而,L系列的销量疲势已经出现,近4个月都未能突破月销4万辆的瓶颈,不及去年下半年的表现。 一季度承压明显,进入四月,理想逐渐出牌。先是在上海车展发布MEGA家庭版车型,并在 5 月 8 日发布了 L 系列的智能焕新版(即L系列的2025 款)。 理想的打法是"加量不加价",除了全系升级智能辅助驾驶硬件,L7/8 是全系标配魔毯双腔空悬、新增 52.3kWh 大电池,L9 则是用上了双枪双阀魔毯空 气悬架、十八点热石按摩等配置。 理想的刀法落得很精确,增配的目的显然是保量。理想手上的牌仅剩理想i8和理想i6,接下来的战,理想要怎么打? 理想L系列,尚能战否? 理想的第一道难关,是要守住增程车型的基本盘。 今年以来,理想L系列车型经历了不同程度的下滑,销量扛把子L6升至回到了去年刚上市的水平,L7的单月销量也都在一万以下。 数据来源: 车主之家 理想L系 ...
理想汽车,压力山大
Hu Xiu· 2025-05-09 10:58
Core Viewpoint - Despite having over 100 billion in cash, Li Auto has not yet reached a point of effortless profitability, facing significant challenges in achieving its sales targets for the year [1]. Group 1: Sales Performance and Targets - Li Auto's sales target for the year is set at 700,000 units, an increase of 200,000 units from the previous year, but the new models, Li i8 and Li i6, are only set to launch in the second half of the year, placing the sales burden primarily on the L series [1][2]. - The L series has shown signs of sales fatigue, failing to exceed 40,000 units per month for the past four months, which is below the performance of the second half of last year [1][3]. - The L series' market share for range-extended products is projected to decline by 2.1% in Q4 2024, indicating increasing competition [6]. Group 2: Competitive Landscape - Li Auto faces growing competition from various manufacturers, including Huawei, which has launched multiple models that directly compete with the L series [6][7]. - The introduction of over 10 new mid-to-large SUVs this year, such as the AITO M8 and Lynk & Co 900, is intensifying competition in the market [7]. - The AITO M8 has already accumulated over 70,000 pre-orders, positioning it as a strong competitor against the L8 and L9 models [7]. Group 3: Product Strategy and Innovations - Li Auto's strategy includes enhancing the L series with new features while maintaining price stability, aiming to preserve sales volume [2][8]. - The L series is undergoing significant upgrades in intelligent driving capabilities, with the introduction of the new Thor-U chip and advanced laser radar systems [8][9]. - The new VLA (Vision Language Action) model aims to differentiate Li Auto's intelligent driving technology from competitors, focusing on advanced capabilities that mimic human-like understanding and reasoning [9][11]. Group 4: Electric Vehicle Strategy - Li Auto's upcoming pure electric models, the i8 and i6, are seen as critical variables for the company's performance this year, with conservative internal sales expectations of 50,000 units for both [15][16]. - The company is exploring new growth avenues through channel transformations and expanding into overseas markets, although the complexity of international markets poses challenges [16][17]. - The domestic market remains the primary focus, but increasing competition and uncertainties in the electric vehicle segment make achieving sales targets difficult [17].