Core Insights - The article discusses a new approach to multi-modal robot learning that addresses the limitations of traditional feature concatenation methods, which often fail in tasks requiring tactile feedback [3][5][33] - The proposed solution involves using compositional policies, where each sensory modality is trained as a separate expert, allowing for more flexible and robust integration of sensory data [9][12][33] Limitations of Current Methods - Traditional multi-modal robot learning typically relies on feature concatenation, which combines all sensor embeddings into a single vector, leading to significant performance drops in tasks requiring tactile information [5][16] - The feature concatenation method treats rare but critical tactile signals as noise, resulting in a drastic decrease in success rates from 35% to 5% when tactile data is added [3][16] Proposed Solutions - The new approach involves training separate expert policies for each modality, allowing for independent learning and reducing interference between modalities [9][12] - This modular design enables easy addition or removal of sensors without the need to retrain the entire system, thus lowering retraining costs and enhancing system robustness [13][16] Performance Results - The proposed method achieved an average success rate of 66% across four RLBench simulation tasks, outperforming single-modal strategies (49%) and feature concatenation (56%) [29] - In specific tasks, the method demonstrated a success rate of 65% for occluded marker picking, compared to 35% for RGB-only and 5% for the concatenation method [34] Robustness and Adaptability - The system shows robustness to runtime disturbances, such as sudden object removal, and can adapt by reallocating weights to remaining functional sensors [21][23] - It maintains stable performance even when simulating sensor failures, demonstrating the effectiveness of the routing network in managing consensus weights [23][27]

Group 1 - The report highlights a significant shift in market sentiment, with the stock market showing strength while the bond market faces pressure, leading to a rapid change in risk appetite [3][13][14] - The yield on the 10-year government bond has risen to 1.75%, while the 30-year bond approaches 2%, indicating a challenging environment for long-duration bonds [3][13] - The report notes that the recent decline in bond prices is characterized by a "local" feature, particularly affecting long-term credit bonds, while short-term credit bonds have shown relative stability [5][48] Group 2 - The report discusses the performance of various bond strategies, indicating that the 30-year government bond strategy has faced the most significant drawdown, with a loss of 192 basis points in the past week [4][21] - Credit strategies have also experienced substantial drawdowns, particularly in bank subordinated bonds and long-duration portfolios [4][21] - Short-term bond strategies have managed to retain some gains from earlier in the year, with certain portfolios even showing positive returns recently [4][21] Group 3 - The report emphasizes that the current bond market environment is marked by a lack of liquidity, particularly in long-term credit bonds, which have seen a sharp decline in trading volume [6][48] - Despite the challenges, the pricing of medium to short-duration credit bonds remains stable, with limited upward movement in yields compared to the adjustments seen at the end of July [6][17] - The report suggests that the stability of the non-bank funding side has contributed to the resilience of short-duration credit bonds [6][70] Group 4 - The report outlines short-term strategies, recommending a cautious approach due to overall low absolute returns [7][71] - It suggests focusing on price spread trading opportunities in bank subordinated bonds and emphasizes the potential for acquiring high-quality city investment bonds with AA+ ratings [7][71] - The report also notes that new credit bond pricing is susceptible to market fluctuations, indicating a need for careful monitoring of market conditions [7][71]